Pedigree Analysis
A woman with two affected brothers asks a genetic counselor a blunt question: what is the chance her son will be affected? Answering it requires more than reading a family tree by eye. It requires turning inheritance patterns into numbers. It also means combining probabilities across generations. Finally, it requires correcting for the ways real biology deviates from Mendel's clean 3:1 world. Pedigree analysis is the discipline that does this. It reconstructs the most likely mode of inheritance from a family history. It assigns genotype probabilities to individuals whose genotypes are not directly observable. And it computes recurrence risk — the probability that a future or existing relative carries or will express a given allele or disorder. This matters because most human genetics happens without controlled crosses. Nobody can self a human being or set up a testcross. Families are the only "experiment," and pedigrees are the record of that experiment as it unfolds across generations. Before molecular diagnostics existed, pedigree analysis was the sole tool for distinguishing autosomal from X-linked, dominant from recessive, and Mendelian from complex traits. It remains the first and cheapest diagnostic step in any genetic counseling session today, even in the genomic era. That is because it tells clinicians which molecular test to order and how urgently. The conceptual leap this article makes beyond simply reading a pedigree is combinatorial. A pedigree is not one Punnett square. It is a chain of conditional probabilities linked across meioses. Bayesian updating — using new information (a later-born unaffected child, a carrier test, an age of onset) to revise a prior probability — is the mathematical engine of modern pedigree analysis. Layered onto this are the complications that real inheritance introduces: reduced Penetrance and Expressivity, variable age of onset, X-inactivation, germline mosaicism, and locus heterogeneity. All of these distort the raw Mendelian ratios that Inheritance Patterns in Pedigrees teaches as a starting template. This article builds from the categorical mode-of-inheritance logic you already know into the quantitative machinery. That machinery includes Bayes' theorem applied to pedigrees and the arithmetic of recurrence risk. It also includes the classic family studies — from Archibald Garrod's alkaptonuria pedigrees to the hemophilia pedigree of the European royal houses. These studies established these inheritance rules as biological fact rather than statistical coincidence.
- Pedigree analysis converts a family history into genotype probabilities and recurrence risks, not just a categorical inheritance mode.
- Bayes' theorem is the core tool: prior probability (from Mendelian expectation) × conditional probability (from new pedigree data) → posterior probability.
- Autosomal recessive, autosomal dominant, X-linked recessive, and X-linked dominant patterns each generate a distinct signature of affected/unaffected transmission across generations.
- Reduced penetrance, variable expressivity, anticipation, mosaicism, and locus heterogeneity all bend the ideal ratios and must be modeled explicitly.
- Garrod's alkaptonuria pedigrees and the hemophilia pedigree of Queen Victoria's descendants are the historical proof that human traits follow Mendelian rules.
- Modern applications: genetic counseling, prenatal risk calculation, carrier screening, and interpreting incidental findings from clinical exome sequencing.
A pedigree encodes a chain of conditional probabilities across meioses, and Bayes' theorem — not a single Punnett square — is what correctly converts that chain into a recurrence risk.
The core idea
A pedigree is a diagrammatic family history: squares for males, circles for females, filled symbols for affected individuals, horizontal lines for matings, vertical lines for offspring, and generations stacked in Roman numerals (I, II, III) with individuals numbered left to right within each generation. Pedigree analysis is the inferential process built on top of that diagram — assigning each unresolved individual a genotype probability distribution consistent with the pattern of affected and unaffected relatives, and then propagating those probabilities forward to compute the risk that a specific, often as-yet-unborn, individual carries or expresses a disease allele. The starting point is always mode of inheritance: is the trait autosomal or X-linked, dominant or recessive, mitochondrial, or does it show a non-Mendelian pattern altogether (multifactorial, imprinted, expanding-repeat)? Each classical mode has a diagnostic signature in a pedigree — for example, autosomal recessive traits typically skip generations and cluster in siblings of unaffected parents, while X-linked recessive traits show affected males connected through unaffected carrier females with no male-to-male transmission. Once a mode is inferred (never with total certainty from a pedigree alone), Mendelian segregation supplies prior probabilities for each possible genotype in each individual: for example, an unaffected child of two known heterozygous carriers of a recessive allele has a 2/3 prior probability of being a heterozygous carrier itself and 1/3 of being homozygous wild-type, since the 1/4 homozygous-affected class is already excluded by observation. The second, and more sophisticated, layer is conditional probability updating. Every additional fact about the pedigree — an unaffected sibling born later, a negative carrier test, an unusually late or absent age of onset, the sex of a fetus — is new evidence that should revise the prior probability via Bayes' theorem. This is where pedigree analysis becomes a genuinely quantitative discipline rather than a pattern-recognition exercise, and where students most often go wrong: treating a pedigree as a single static Punnett square rather than as a Bayesian network of dependent probabilities threaded through real meioses. The complications layered on top — penetrance less than 100%, age-dependent expressivity, anticipation in trinucleotide-repeat disorders, mitochondrial heteroplasmy, and locus heterogeneity (the same phenotype produced by mutations at different loci, distinguishable in principle by Complementation Analysis) — must each be built explicitly into the probability model, because ignoring them silently converts a correct recurrence-risk calculation into a wrong one.
Mechanism
Step 1 — Establish the mode of inheritance from pattern features. Five signatures are diagnostic, though none is individually proof:
| Mode | Key pedigree signature |
|---|---|
| Autosomal recessive | Skips generations; affected often from two unaffected (carrier) parents; equal sex ratio; consanguinity raises risk |
| Autosomal dominant | Affected in every generation; affected × unaffected → ~1/2 affected offspring; no skipping (barring nonpenetrance) |
| X-linked recessive | Mostly affected males; transmitted through unaffected carrier mothers; no male-to-male transmission; affected father's daughters are all carriers |
| X-linked dominant | Affected females roughly 2× affected males; affected father → all daughters affected, no sons affected |
| Mitochondrial | Transmitted only through affected mothers to all offspring; no paternal transmission; variable severity from heteroplasmy |
Step 2 — Assign genotype priors using Mendelian segregation. Once mode is fixed, meiosis (independent segregation of homologs, and for X-linked traits, segregation of the sex chromosomes) fixes the prior probabilities of genotypes among offspring of parents with known or inferred genotypes. This is ordinary Mendelian ratio arithmetic, but applied person-by-person rather than trait-by-trait.
Step 3 — Update priors with conditional (Bayesian) evidence. Any observation whose probability differs between the hypotheses being compared updates the prior. The general form: P(H | data) = [P(data | H) × P(H)] / Σ [P(data | Hᵢ) × P(Hᵢ)] In pedigree work, H is usually "individual X is a carrier/is affected/will develop disease" and the data are things like "individual X has had two unaffected sons" or "individual X tested negative for the family's known mutation."
Step 4 — Correct for penetrance, expressivity, and heterogeneity. If penetrance is f (0 ≤ f ≤ 1), then P(affected | genotype = disease allele present) = f, not 1; an unaffected individual carrying the disease genotype occurs with probability (1 − f), and this must be folded into every likelihood term. Age-dependent penetrance (as in Huntington disease or BRCA1-associated breast cancer) requires penetrance to be expressed as a function of current age, f(age), further conditioning the calculation on the individual's age at assessment.
Step 5 — Propagate risk to the individual of interest. The posterior genotype probability of a parent is combined with the Mendelian segregation probability for the specific relationship to compute the final recurrence risk for a child, sibling, or more distant relative — multiplying probabilities down independent branches of the pedigree and summing over mutually exclusive paths.
The evidence
Garrod's alkaptonuria pedigrees (1902). Archibald Garrod, a physician studying the black-urine disease alkaptonuria, noticed that the disorder appeared disproportionately in the offspring of first-cousin marriages and often affected multiple siblings whose parents were both unaffected — a pattern that skipped generations exactly as recessive inheritance predicts. Garrod consulted William Bateson, who had just reintroduced Mendel's work to English-speaking science, and Bateson recognized the pedigree pattern as a textbook autosomal recessive segregation: unaffected heterozygous parents, roughly one-quarter of offspring affected, no bias in sex ratio, and enrichment among consanguineous unions (which raise the chance both parents carry the same rare recessive allele identical by descent). This was the first documented case of a human trait shown, by pedigree logic alone, to obey Mendelian segregation, and it founded the field of inborn errors of metabolism — Garrod correctly inferred that alkaptonuria resulted from a missing enzymatic step (later identified as homogentisate 1,2-dioxygenase deficiency), decades before biochemical genetics existed as a discipline.
The hemophilia pedigree of European royalty (19th–20th century). Queen Victoria of the United Kingdom (1819–1901) carried a spontaneous mutation in the F9 gene (encoding coagulation factor IX, causing hemophilia B) on one of her two X chromosomes — a mutation that arose either in one of her parents' germlines or in her own early embryogenesis, since neither of her parents nor her extensive ancestry shows evidence of the disease. Through dynastic marriages, her carrier daughters and granddaughters transmitted the allele into the royal houses of Russia, Spain, and Germany. The resulting pedigree — extensively documented because European royal medical records were kept meticulously — shows the unmistakable X-linked recessive signature: affected sons of carrier mothers (most famously Tsarevich Alexei Nikolaevich of Russia), no affected daughters, no male-to-male transmission, and transmission skipping through phenotypically normal carrier females. Molecular sequencing of exhumed royal remains in 2009 confirmed the exact causal mutation and its X-linked segregation, validating a century of pedigree-based inference with direct molecular evidence.
Modern statistical validation — the Elston-Stewart and Lander-Green algorithms. The intuitive Bayesian logic used by hand in classical pedigrees was formalized into rigorous computational algorithms: Elston and Stewart (1971) developed a peeling algorithm for computing exact likelihoods on pedigrees of arbitrary size, and Lander and Green (1987) developed a hidden Markov model approach for multipoint linkage analysis across many markers. These algorithms, still used in clinical linkage and genetic-risk software today, are direct mathematical descendants of the same conditional-probability logic taught at the bedside — the historical pedigree case studies (Garrod's alkaptonuria families, the royal hemophilia pedigree, and thousands of subsequent Mendelian disease pedigrees collected in databases such as OMIM) are the empirical foundation that these algorithms were built and validated against.
Worked examples
Example 1 — Bayesian recurrence risk in X-linked recessive disease with a nonpenetrant complication removed. A woman, Ann, has a brother affected with Duchenne muscular dystrophy (X-linked recessive, essentially fully penetrant in males, DMD gene). Her mother must be an obligate carrier (Xᴬ Xᵃ) because the son is affected and the father is unaffected and non-carrier by definition (males are hemizygous). Ann herself is unaffected. What is the probability Ann is a carrier, and what is the risk that Ann's son is affected — first with no further information, then after Ann has already had one unaffected son?
- Prior: Ann's mother is Xᴬ Xᵃ, so Ann has a 1/2 prior probability of inheriting Xᵃ (carrier) and 1/2 of inheriting Xᴬ (non-carrier). She is phenotypically unaffected in either case (carriers of DMD are usually unaffected or mildly affected due to skewed X-inactivation, but let's treat this idealized case as fully unaffected regardless of genotype, i.e., the observation "Ann is unaffected" carries no information).
- If Ann is a carrier (P = 1/2): each son has 1/2 chance of inheriting Xᵃ from her → 1/2 chance affected.
- If Ann is a non-carrier (P = 1/2): each son has 0 chance of being affected.
- Unconditional prior risk that Ann's son is affected = (1/2 × 1/2) + (1/2 × 0) = 1/4.
Now suppose Ann has already had one son, and he is unaffected. This is new Bayesian evidence.
- P(unaffected son | Ann carrier) = 1/2
- P(unaffected son | Ann non-carrier) = 1
- P(Ann carrier | unaffected son) = [1/2 × 1/2] / [1/2 × 1/2 + 1/2 × 1] = (1/4) / (3/4) = 1/3
- P(Ann non-carrier | unaffected son) = 2/3
Updated risk for a subsequent son = (1/3 × 1/2) + (2/3 × 0) = 1/6. This is the classic demonstration that a single additional unaffected child meaningfully lowers recurrence risk — from 1/4 down to 1/6 — because "unaffected son" is three times more probable if Ann is a non-carrier than if she is a carrier, exactly the asymmetry Bayes' theorem is built to exploit.
Example 2 — Autosomal dominant disorder with reduced, age-dependent penetrance. Consider a family with a late-onset autosomal dominant neurodegenerative disorder in which penetrance by age 60 is 80% (f = 0.8) — i.e., 20% of gene carriers remain unaffected even at age 60 or beyond (nonpenetrant carriers). A 55-year-old woman, Beth, is currently unaffected. Her father was affected (and therefore an obligate heterozygote, Aa), so Beth's Mendelian prior probability of having inherited the disease allele is 1/2. What is the probability Beth actually carries the allele, given that she is currently unaffected at age 55, and what is the risk to her children?
- P(unaffected at 55 | carrier) ≈ 1 − f(55) — suppose penetrance at age 55 specifically is 0.65 (lower than the age-60 lifetime estimate, since penetrance rises with age) → P(unaffected | carrier) = 0.35
- P(unaffected at 55 | non-carrier) = 1 (non-carriers never show disease)
- Prior P(carrier) = 1/2, P(non-carrier) = 1/2
- P(carrier | unaffected at 55) = [0.5 × 0.35] / [0.5 × 0.35 + 0.5 × 1] = 0.175 / 0.675 ≈ 0.26
Beth's risk of carrying the mutation drops from the naive Mendelian 50% to about 26% once her current unaffected status at age 55 is incorporated — but it does not drop to zero, precisely because penetrance is incomplete. The risk to each of Beth's children of inheriting the allele = 0.26 × 1/2 ≈ 13%, dramatically lower than the naive (and wrong) calculation of 1/2 × 1/2 = 25% that ignores penetrance entirely. This example demonstrates why reduced penetrance must be built into the likelihood term, not bolted on afterward — a common and clinically consequential error.
Common misconceptions & traps
| Misconception | Correction |
|---|---|
| "A pedigree alone can prove the exact mode of inheritance." | Small pedigrees are often consistent with more than one mode (e.g., a rare autosomal dominant with low penetrance can mimic recessive skipping); mode is inferred, not proven, until molecular data confirm it. |
| "Each birth is independent of prior information, so risk stays at the Mendelian ratio no matter what." | Bayesian updating is essential: additional unaffected relatives, negative test results, or advanced unaffected age all revise posterior risk, sometimes substantially. |
| "A recurrence risk of 1/4 means exactly one in four children will be affected." | Each pregnancy is an independent trial with probability 1/4; it does not mean 1 in exactly 4 actual children will be affected, especially in small sibships. |
| "Dominant means common in the population; recessive means rare." | Dominance describes the genotype-phenotype relationship within an individual, not allele frequency in a population — dominant disease alleles can be extremely rare. |
| "An unaffected carrier of a dominant disease allele contradicts dominance." | This is exactly what reduced penetrance describes — the genotype is dominant in its mode of inheritance, but expression of the phenotype fails in some carriers. |
| "Two affected parents guarantee affected offspring for a dominant trait." | Only if both parents are homozygous; if either is heterozygous, 1/4 to 1/2 of offspring can be unaffected depending on cross type. |
The scientist behind it
Archibald Edward Garrod (1857–1936) was a London physician who, unusually for his era, treated inherited metabolic disease as a problem for chemistry and family history simultaneously, rather than as separate clinical curiosities. Trained at Oxford and St Bartholomew's Hospital, Garrod spent years cataloguing patients with alkaptonuria, whose urine turns black on exposure to air. Rather than dismiss the clustering of cases within certain families as coincidence, he systematically recorded which relatives were affected and noted the striking excess of first-cousin marriages among the parents of affected children — a pattern he could not explain until a colleague, the biologist William Bateson, pointed him to Mendel's recently rediscovered ratios. Garrod's 1902 paper "The Incidence of Alkaptonuria: A Study in Chemical Individuality" is now recognized as the founding document of both human biochemical genetics and evidence-based pedigree analysis, though it drew little attention at the time and was only fully appreciated decades later as biochemistry and genetics matured into a unified discipline.
Origins · 1 minArchibald Edward Garrod and how this was discovered
Where this matters today
Genetic counseling clinics use quantitative pedigree analysis as the first and most cost-effective diagnostic step before ordering expensive sequencing, calculating a couple's recurrence risk for conditions ranging from cystic fibrosis to hereditary breast and ovarian cancer (BRCA1/BRCA2) before any laboratory test is run. Prenatal risk calculations for X-linked disorders such as Duchenne muscular dystrophy or hemophilia routinely use exactly the Bayesian carrier-probability method shown above, updated in real time as new family members are tested or born. Cancer genetics clinics apply penetrance-corrected pedigree models (such as BRCAPRO and the Claus model) to estimate an individual's probability of carrying a BRCA1/BRCA2 mutation from family cancer history alone, guiding decisions about genetic testing, enhanced screening, or prophylactic surgery. In agriculture and animal breeding, pedigree-based BLUP (best linear unbiased prediction) methods extend the same conditional-probability logic across large livestock pedigrees to estimate breeding values for traits like milk yield or disease resistance. In conservation biology, pedigree reconstruction from molecular markers in small, endangered populations is used to minimize inbreeding depression in captive breeding programs, applying the same recessive-allele logic Garrod used a century earlier, now at population scale. Clinical exome and genome sequencing increasingly returns incidental variants of uncertain significance, and pedigree segregation analysis — checking whether a candidate variant tracks with disease status across the family exactly as the proposed inheritance mode predicts — remains one of the strongest lines of evidence for classifying a variant as pathogenic versus benign.
Ask the tutor: How would the recurrence risk calculation change if the disorder in Example 1 showed germline mosaicism in one parent rather than simple X-linked transmission?
Why isn't a pedigree ever fully proof of a specific mode of inheritance?
Small family sizes make many modes statistically indistinguishable: a rare autosomal dominant condition with markedly reduced penetrance can look exactly like autosomal recessive skipping, and X-linked recessive traits in families with few male births can be mistaken for autosomal recessive. Pedigree analysis produces the most probable mode given the observed pattern, formally comparable using likelihood ratios across competing models, but molecular confirmation (sequencing the causal gene, linkage analysis, or segregation of a known variant) is required for certainty. This is precisely why clinical genetics treats pedigree-derived inheritance mode as a strong hypothesis to guide testing, not a final diagnosis.
How does reduced penetrance formally enter a Bayesian pedigree calculation?
Penetrance, f, is the probability that a given genotype produces the expected phenotype. It modifies the likelihood term P(observed phenotype | genotype): for a dominant disease allele with penetrance f, P(affected | Aa) = f and P(unaffected | Aa) = 1 − f, whereas P(unaffected | aa) = 1 exactly. These likelihoods are multiplied against the Mendelian genotype prior and normalized, exactly as in Example 2. Ignoring penetrance sets f = 1 implicitly, silently overstating how informative an unaffected phenotype really is about underlying genotype.
What is locus heterogeneity, and why does it complicate pedigree-based risk calculation?
Locus heterogeneity means that the same or a very similar phenotype (for instance, retinitis pigmentosa or nonsyndromic deafness) can be caused by mutations at different, unlinked genes in different families. Within a single family this rarely matters, because the segregating mutation is presumably fixed at one locus, but it means recurrence risk figures and even the inferred inheritance mode from one family cannot automatically be applied to an unrelated family with the "same" diagnosis. Complementation Analysis was historically used to test whether two mutant lines (or, in principle, two human families) fail to complement, indicating mutation at the same locus, distinguishing true allelism from phenocopies caused by different genes.
Why did the hemophilia pedigree of European royalty become such an important historical case study?
Royal families kept unusually complete, cross-referenced medical and genealogical records across many nations and generations, producing an exceptionally large, well-documented pedigree for a rare X-linked disease. It showed, unambiguously and at scale, every diagnostic feature of X-linked recessive inheritance: affected sons of unaffected carrier mothers, complete absence of male-to-male transmission, and transmission "skipping" visibly through phenotypically normal carrier daughters across three generations spanning Britain, Russia, Germany, and Spain. When DNA extracted from the remains of the Russian imperial family in 2009 confirmed the exact F9 mutation, it validated over a century of purely pedigree-based inference with direct molecular evidence.
How is anticipation handled differently from ordinary penetrance in pedigree analysis?
Anticipation is the phenomenon, seen in trinucleotide-repeat disorders like Huntington disease and myotonic dystrophy, where age of onset decreases and/or severity increases in successive generations because repeat expansions tend to grow during meiosis (especially paternal meiosis in Huntington disease). Unlike ordinary penetrance, which is treated as a fixed probability, anticipation requires the pedigree model to track repeat length or generation number as an additional variable influencing age-dependent penetrance, since a fixed single penetrance value would systematically misestimate risk for children versus grandchildren of an affected founder.
Can pedigree analysis distinguish autosomal dominant inheritance from a multifactorial (polygenic/environmental) trait?
Yes, though imperfectly, using pattern and recurrence-risk shape rather than a single feature. Mendelian dominant traits show recurrence risk that stays roughly constant (about 1/2) regardless of how many relatives are affected, whereas multifactorial traits show recurrence risk that rises with the number of affected relatives and with the closeness of relationship, and drops off faster than Mendelian expectation in more distant relatives. Twin studies (comparing monozygotic versus dizygotic concordance) and formal segregation analysis are typically needed alongside the pedigree to make this distinction with confidence.
Why do genetic counselors still draw pedigrees by hand when whole-genome sequencing is available?
A pedigree costs nothing, takes twenty minutes, and immediately narrows which genes and inheritance modes are plausible, focusing an otherwise enormous sequencing search space and helping interpret variants of uncertain significance through segregation with disease status in the family. It also captures information sequencing cannot, such as consanguinity, reproductive history, and stillbirths, and it is the tool used to explain risk to patients in terms they can see and reason about, which remains central to informed reproductive decision-making regardless of how sophisticated the underlying molecular technology becomes.