Contents
14 — Linkage, recombination and mapping Part 2 — Transmission genetics

14 — Linkage, recombination and mapping

Before this: Ch 09 · Ch 10 · Ch 12 · Time: ~50 min

Statistics needed: S2 Distributions

What you'll be able to do

The core idea

is a fact about , not about genes. Two loci assort independently because they sit on different chromosomes, and the chromosomes line up independently at metaphase I. Put two loci on the same chromosome and that mechanism is gone: they are physically tethered, and they travel to the same gamete unless something cuts the tether between them.

The thing that cuts the tether is a crossover. Crossovers happen at roughly uniform density along the chromosome, so the probability that a crossover falls between two loci grows with the distance between them. Measure how often the tether breaks and you have measured a distance — without knowing what a chromosome is made of, without seeing it, without any molecular biology whatsoever. This is how the first genome maps were built, two decades before anyone knew DNA was the genetic material.

The catch is that the ruler saturates. Beyond a certain separation the tether breaks so reliably that two loci on the same chromosome become statistically indistinguishable from two loci on different chromosomes. The measurement stops scaling. Everything about how genetic maps are actually constructed is a workaround for that ceiling.


1. Linkage is a violation of independent assortment

Recall the mechanism (Ch 09): each orients at random on the metaphase I plate, independently of every other bivalent. That randomness is the physical source of Mendel's second law. It applies to loci on different and to nothing else.

Two loci on the same chromosome are syntenic. If they are close enough that a crossover rarely separates them, they are linked — a stronger claim, and a measurable one. Loci at opposite ends of a large chromosome are syntenic but not linked in any detectable sense.

Consider a doubly heterozygous individual. Which are on which physical chromosome — the — determines what "recombinant" means:

   coupling  (cis)                          repulsion  (trans)

   ── A ————————— B ──  from one parent     ── A ————————— b ──
   ── a ————————— b ──  from the other      ── a ————————— B ──

   parental gametes:    A B  and  a b       parental gametes:    A b  and  a B
   recombinant gametes: A b  and  a B       recombinant gametes: A B  and  a b

The four gamete classes are the same in both cases. The labels are opposite. Recombinant is not a property of a gamete; it is a property of a gamete relative to the chromosomes that entered the . Miss this and every calculation in the chapter inverts.

This is why the testcross — heterozygous parent crossed to a homozygous tester — is the workhorse design. The tester contributes only recessive alleles, so each offspring's is a direct readout of the heterozygous parent's gamete: no inference, no dominance masking anything. You are counting gametes.

2. Recombination frequency, and what a centimorgan is

\text{RF} = \frac{\text{number of recombinant gametes}}{\text{total gametes}}

For unlinked loci, RF = 0.5: the four gamete classes appear in equal numbers, and "parental" and "recombinant" are labels on an even split. For tightly linked loci, RF ≈ 0.

Sturtevant's insight in 1913 — as an undergraduate in Morgan's lab, reportedly in a single night, using six Drosophila genes — was that if crossovers occur at roughly uniform density, then RF is monotone in physical separation, so RF values can be used to order loci along a line and to space them. The first genetic map was a one-dimensional arrangement inferred entirely from breeding statistics.

The unit is defined to make the bookkeeping trivial:

1 map unit = 1 (cM) = 1% recombination, for small intervals.

The hedge matters. The centimorgan is properly defined as one hundredth of a Morgan, and a Morgan is the map length over which one crossover is expected per gamete. So:

These coincide when d is small, because two crossovers in a tiny interval is a second-order event. They diverge badly when d is large. Keeping d and r mentally separate is most of what makes this chapter tractable.

3. The 50% ceiling, derived twice

Argument one — four strands. By the time happens, each has replicated, so a contains four chromatids. A single crossover involves exactly two of them:

   chromatid 1  ─────────────────────   A ····· B     parental
   chromatid 2  ─────────╳───────────   A ····· b     recombinant
   chromatid 3  ─────────╳───────────   a ····· B     recombinant
   chromatid 4  ─────────────────────   a ····· b     parental

One crossover between A and B therefore yields 2 recombinant and 2 parental products from that : exactly 50%. Not 100%. A meiosis in which a crossover is guaranteed to fall between two loci still produces only half recombinant gametes, and no number of additional crossovers can push the average above that.

Statistics: the Poisson distribution — including the row where its parameter is a chromosome's map length in Morgans — is covered in S2 §2.

Argument two — parity. Model crossovers along the interval as a Poisson process, and let K be the number affecting a given chromatid, with mean d. A chromatid is recombinant across the interval if and only if K is odd — two crossovers put the flanking markers back where they started. So

r = P(K \text{ odd}) = e^{-d}\left(d + \frac{d^3}{3!} + \frac{d^5}{5!} + \cdots\right) = e^{-d}\sinh d = \frac{1 - e^{-2d}}{2}

As d \to \infty, r \to 1/2 from below. As d \to 0, r \to d. Both limits are the ones the biology demands, and we have just derived the Haldane mapping function on the way past — more on that in §7.

RF = 50% does not mean "50 cM apart". It means the ruler has saturated and returned no information. Two loci at opposite ends of chromosome 1 (~280 cM) and two loci on different chromosomes give the identical observation. A genetic map is therefore never measured directly at long range: it is built additively out of short intervals, each one measured against markers close enough that multiple crossovers are rare, and then chained. Every long genetic distance you have ever seen quoted is a sum, not a measurement.

4. Why two-point crosses are not enough

A two-point cross gives you one number per pair of loci. Three loci give three pairwise RFs, and they will not add up — the outer distance always comes out short, because double crossovers between the outer pair restore the parental configuration and are silently counted as non-recombinant:

   physical order:   A ————— region I ————— B ————— region II ————— C

   no crossover          A  B  C        parental at every pair
                         a  b  c

   crossover in I        A  b  c        recombinant A–B  and  A–C
                         a  B  C

   crossover in II       A  B  c        recombinant B–C  and  A–C
                         a  b  C

   crossover in I & II   A  b  C   ←──  A and C are back in the parental
                         a  B  c        combination. Invisible to A–C.
                                        Only B has changed partners.

Two-point data also cannot establish order. RF(A,C) being the largest of three is suggestive, but with three roughly similar numbers and sampling noise it is not decisive, and it gives you no way to detect the double crossovers that caused the shortfall.

The three-point cross fixes both problems at once, because that last row of the diagram is also the signature that identifies the middle locus.

5. The three-point testcross

Cross a triply heterozygous parent to a triply homozygous tester. Score the progeny. There are 2^3 = 8 classes, in four reciprocal pairs, and the pairs sort themselves by frequency:

graph TD A["8 phenotype classes<br/>sorted by count"] --> B["Two <b>most</b> frequent<br/>= parental (non-crossover)<br/>→ reveals the F1's phase"] A --> C["Two <b>least</b> frequent<br/>= double crossover<br/>→ P(both) = product of two small numbers"] B --> D["Compare a DCO class to a parental class.<br/><b>Exactly one locus differs.</b><br/>That locus is in the middle."] C --> D D --> E["Now the order is known:<br/>assign the two remaining pairs<br/>to region I or region II"] E --> F["RF(I) = (SCO<sub>I</sub> + DCO)/N<br/>RF(II) = (SCO<sub>II</sub> + DCO)/N<br/><i>double crossovers count in both</i>"] F --> G["c.o.c. = DCO<sub>obs</sub> / (RF<sub>I</sub> × RF<sub>II</sub> × N)<br/>interference = 1 − c.o.c."]

The middle-locus rule is not a trick. A double crossover puts one exchange on each side of the middle marker, physically swapping that marker onto the other chromatid while leaving both flanks where they were. The DCO class is the parental class with the middle locus flipped, so reading off which locus flipped reads off the order.

Double crossovers count in both intervals, because a DCO gamete experienced an exchange in region I and one in region II. Omitting them undercounts both distances — the single most common arithmetic error in transmission genetics. Even so, the corrected interval distances remain slight underestimates: a double crossover within region I is still invisible. That residual bias is what mapping functions correct.

6. Tetrad analysis: when you can see all four products

Every design so far counts gametes: one product per , dropped into a pool of millions, and you rebuild the meiosis statistically from the aggregate. Fungi let you skip the rebuilding.

In Saccharomyces cerevisiae and Neurospora crassa the four haploid products of one meiosis stay together inside a sac, the ascus. Dissect it, plate the spores separately, and you have genotyped a single meiosis completely. The exact 2:2 that Ch 09 §7 called deterministic stops being an inference.

Yeast asci are unordered — four spores loose in a bag. Neurospora asci are ordered: the ascus is barely wider than a spore, so the meiosis I and II spindles cannot slip past one another, and the row of spores records which product went to which pole — strictly more information, spent in the last two subsections.

Unordered tetrads: PD, NPD, TT

Cross A B × a b and dissect. Only three ascus compositions exist:

   parental ditype         non-parental ditype          tetratype
        (PD)                      (NPD)                    (TT)

       A B                        A b                      A B
       A B                        A b                      A b
       a b                        a B                      a B
       a b                        a B                      a b

   2 genotypes,              2 genotypes,             4 genotypes,
   both parental             both recombinant         2 parental, 2 recombinant

Derive where each comes from, using nothing but the four-strand picture of §3.

No crossover in the interval → PD. Nothing moved.

One crossover → TT, always. A crossover joins two non-sister chromatids, so exactly two of the four products are recombinant. That is the 50% ceiling of §3, seen one meiosis at a time.

Two crossovers → it depends which chromatids they used. There are four non-sister pairs; the second exchange reuses the first pair, shares one chromatid with it, or takes the other entirely:

   double crossover    chromatids used      tetrad    share
   ────────────────    ───────────────      ──────    ─────
   2-strand            1–3, then 1–3          PD       1/4
   3-strand (a)        1–3, then 1–4          TT       1/4
   3-strand (b)        1–3, then 2–3          TT       1/4
   4-strand            1–3, then 2–4         NPD       1/4

so doubles split 1 PD : 2 TT : 1 NPD. That 1:2:1 says there is no chromatid — a different claim from the crossover interference of §7, and empirically close to true.

NPD therefore has exactly one source: a four-strand double crossover. Both exchanges must land in the interval and use all four chromatids. It is the tetrad analogue of the three-point cross's DCO class: rare, and for that reason the most informative row in the table.

ClassArises fromRecombinant chromatidsBetween linked loci
PDno crossover, or a 2-strand double0 of 4the bulk of tetrads
TTa single crossover, or a 3-strand double2 of 4intermediate
NPDa 4-strand double, and nothing else4 of 4rare, and second-order in d

The linkage test: compare PD with NPD

For loci on different , PD and NPD come from the same coin: the two bi-orient independently at metaphase I, so whichever way the second faces, the tetrad reads PD or NPD with probability ½ each. Tetratypes come from elsewhere — a crossover between either locus and its own — so TT carries no information at all.

PD ≈ NPD ⟹ unlinked. PD ≫ NPD ⟹ linked.

Throw away TT and what remains is a binomial test of a 1:1 ratio, which behaves far better than asking whether an observed RF sits below 0.5.

The mapping formula

Count recombinant chromatids directly. Each PD contributes 0 of 4, each TT 2 of 4, each NPD 4 of 4:

\text{RF} = \frac{\text{NPD} + \tfrac{1}{2}\,\text{TT}}{N}

Correct as a , and defective for the same reason a two-point cross is: the doubles you can see (NPD) prove there were doubles you cannot, and those went uncounted.

Fix it by counting crossovers instead. Let S be the tetrads with one crossover in the interval and D those with a double. Every single gives TT; every double splits 1:2:1. So

\text{TT} = S + \tfrac{1}{2}D \;\Rightarrow\; S = \text{TT} - 2\,\text{NPD}

Every NPD tetrad you see stands for three double-crossover tetrads you cannot. Total crossovers = S + 2D = (\text{TT} - 2\,\text{NPD}) + 8\,\text{NPD} = \text{TT} + 6\,\text{NPD}, and that is crossovers per tetrad. Map distance is crossovers per chromatid, and each crossover uses 2 of the 4 — the same halving as §3 — so

\boxed{\;d = \frac{\tfrac{1}{2}\,\text{TT} + 3\,\text{NPD}}{N}\ \text{Morgans}\;}

the Perkins equation, usually printed as 100(\text{TT} + 6\,\text{NPD})/2N cM. The ½ is the four-strand factor. The 3 is 6/2, and the 6 is the 8 crossovers carried by the 4\,\text{NPD} double-crossover tetrads, less the 2\,\text{NPD} of them already counted once inside TT.

What comes out is a map distance, not a recombination frequency (§2), and it is not bounded by 0.5. It corrects for double crossovers inside the interval — the residual bias §5 could only , and what §7 attacks from the other side. The difference is the evidence: a mapping function assumes a crossover distribution and infers the hidden doubles; Perkins counts them, because NPD makes a quarter of them visible.

Worked. 200 yeast tetrads from a two-marker cross: PD = 130, NPD = 4, TT = 66.

   linkage test    130 vs 4 against 1:1          binomial p ≈ 6 × 10⁻³⁴   → linked
   naive RF        (4 + 66/2) / 200   = 0.185    → 18.5%
   Perkins         (66/2 + 3×4) / 200 = 0.225 M  → 22.5 cM
   recovered       D = 4 × 4 = 16 doubles,  S = 66 − 8 = 58 singles,
                   130 − 4 = 126 tetrads with no crossover;  126 + 58 + 16 = 200 ✓

Hand the naive 0.185 to the §7 mapping functions and Haldane returns 23.1 cM, Kosambi 19.4 cM. Perkins' 22.5 lands beside Haldane, and should: both assume crossovers are independent.

Ordered tetrads: distance to the centromere

Now spend the spore order. Score one locus and ask only when its two separated:

   no crossover between locus              a crossover between them
   and centromere
   → alleles part at meiosis I             → alleles part at meiosis II

   FIRST-DIVISION SEGREGATION              SECOND-DIVISION SEGREGATION

      A A A A a a a a                         A A a a A A a a
      └──── 4 : 4 ────┘                       └── 2 : 2 : 2 : 2 ──┘

                                           (or 2 : 4 : 2 — the block of four
                                            A spores is broken either way)

Eight spores, not four: a post-meiotic mitosis doubles each product and the pair stays adjacent.

A crossover between the locus and the centromere joins two non-sister chromatids, so two of the four products end up attached to the centromere they did not start with. A second-division ascus is half recombinant, exactly as in §3. Convert asci to chromatids by halving:

d_{\text{centromere}} = \frac{1}{2}\cdot\frac{\text{second-division asci}}{\text{total asci}}

24 second-division asci in 100 gives ½ × 0.24 = 0.12 M = 12 cM to the centromere. The measurement saturates at ⅔ second-division asci and therefore at 33.3 cM — the ceiling of §3 in different clothing.

This is the only classical method that measures a distance to something you cannot mutate. There is no centromere allele to score; the ordered ascus reports the position of a structure.

Gene conversion, observed rather than inferred

2:2 is not a statistical expectation. One allele per homolog, replicated once, four products — the ratio is mechanically forced. So an ascus reading 3:1 (6:2 among the eight spores) states that one allele was overwritten by the other, and 5:3 states that one product carried an unrepaired mismatch through the post-meiotic mitosis, leaving its two spores disagreeing. Non-reciprocal transfer, on the bench, in a count.

None of it survives pooling: a 3:1 ascus and a 1:3 ascus cancel in a gamete count, which is why stayed invisible until someone kept the four products together. Ch 18 §5 takes these ratios as its evidence and supplies the heteroduplex and behind them.

For the programmer

A testcross is a sampled aggregate over runs: one output per run, and you recover the distribution from the marginals. A tetrad is the complete output of one run — the joint distribution, not the marginal.

That is why fewer meioses buy more information. A four-strand double crossover emits four recombinant chromatids, and loose in a gamete pool they look like four ordinary ones; the event is identifiable only because you know they came from the same run. Same data volume, strictly more structure — and the structure is where the doubles, the centromere and the conversions live.

7. Interference and mapping functions

If crossovers were independent along the , the expected number of double crossovers would be N \times \text{RF}_\text{I} \times \text{RF}_\text{II}. They are not independent. Define

\text{c.o.c.} = \frac{\text{observed DCO}}{\text{expected DCO}}, \qquad I = 1 - \text{c.o.c.}

Almost universally I > 0: positive , a crossover suppressing others nearby. It is strong and long-range — in humans the inhibition extends over tens of megabases, which is why most chromosome arms receive close to one crossover and rarely more, and why the obligate crossover (at least one per , needed for correct ) is a the cell can nearly saturate. The mechanism belongs to Ch 18; here it is a parameter.

Now turn that parameter into a mapping function. Take an interval of map length d with recombination fraction r, and extend it by an infinitesimal \delta. The extended interval is recombinant iff exactly one of the two pieces is. Let c be the coefficient of coincidence between the two pieces, so that P(\text{rec in } \delta \mid \text{rec in } d) = c\,\delta and hence P(\text{rec in } \delta \wedge \text{not rec in } d) = \delta - c r \delta:

r(d+\delta) = \underbrace{r\,(1 - c\delta)}_{\text{rec in } d \text{ only}} + \underbrace{(\delta - c r \delta)}_{\text{rec in } \delta \text{ only}} = r + \delta\,(1 - 2cr)

Taking \delta \to 0:

\boxed{\;\frac{dr}{dd} = 1 - 2\,c(r)\,r\;}

Every mapping function in use is a choice of c(r):

FunctionAssumptionc(r)ODESolution
Morgancomplete interference0r' = 1r = d — only valid for tiny d
Haldane (1919)no interference; Poisson crossovers1r' = 1 - 2rr = \tfrac{1}{2}(1 - e^{-2d}),   d = -\tfrac{1}{2}\ln(1-2r)
Kosambi (1944)interference decaying with distance2rr' = 1 - 4r^2r = \tfrac{1}{2}\tanh(2d),   d = \tfrac{1}{4}\ln\!\frac{1+2r}{1-2r}

Haldane's c = 1 is the same "crossovers are independent" assumption that produced the Poisson derivation in §3 — the two routes agree, as they must. Kosambi's c = 2r is an empirical fit with the right shape: interference is complete for adjacent loci (r \to 0 \Rightarrow c \to 0) and absent for distant ones (r \to 0.5 \Rightarrow c \to 1). It has no mechanistic derivation and does not need one.

Both are strictly increasing, both send r = 0.5 to d = \infty, and both are near-identities below about 10 cM. Above that Haldane inflates distances, because it credits the data with hidden double crossovers that interference actually prevented. In organisms with real interference — which is most of them — Kosambi is the better default.

8. Genetic distance is not physical distance

They are different rulers measuring different things: a is a probability of exchange, a megabase is a count of nucleotides. The conversion between them is a local property of the genome, and treating it as a constant is one of the more expensive habits in applied genomics.

QuantityValue
Human sex-averaged autosomal map~3,400–3,500 cM
Male (paternal) map~2,600–2,700 cM · ~26 crossovers per sperm
Female (maternal) map~4,200–4,400 cM · ~42 crossovers per egg
Genome-wide average rate~1.2 cM/Mb, i.e. 1 cM ≈ 0.8–1 Mb
Hotspot width~1–2 kb
Hotspots detected genome-wide~30,000–50,000
Share of recombination in hotspots~80% of events in <15% of the sequence

The "crossovers per gamete" column is not an extra fact — it is the definition of the Morgan applied to the whole genome. A 2,600 cM map is 26 Morgans is 26 expected crossovers per gamete. Cytological counts agree once you correct for strandedness: about 50 crossover sites are visible per human spermatocyte, and each involves 2 of 4 chromatids, so each gamete inherits about half of them.

The non-uniformity is severe, and structured:

So: use a genetic map, not a multiplier. Averaged over the genome 1 cM ≈ 1 Mb; at any particular locus that estimate can be wrong by more than an order of magnitude either way.

9. LOD scores: linkage analysis in humans

Humans do not permit testcrosses, family sizes are small, and is frequently unknown. The response is to stop counting recombinants and start comparing likelihoods.

Statistics: everything below is derived here from the binomial alone (S2 §1), so you do not need the likelihood track yet. When you reach it, S6 §§1–4 recomputes this same in code and puts it in its general setting.

For a recombination fraction \theta, define

Z(\theta) = \log_{10}\frac{L(\theta)}{L(0.5)}

— the base-10 log of the odds that the observed inheritance pattern arose under at \theta versus under no linkage. For n fully informative, phase-known meioses of which k are recombinant, the likelihood is binomial and the constant cancels:

Z(\theta) = \log_{10}\frac{\theta^k (1-\theta)^{n-k}}{0.5^{\,n}}, \qquad \hat\theta = \frac{k}{n}

Two properties make this workable.

LOD scores add across independent families. Log-likelihoods add, so a laboratory could publish a table of Z(\theta) for its and another laboratory could sum it with its own. In the decades before cheap computing, this additivity is what made human linkage analysis a collective enterprise at all.

Statistics: combining a likelihood ratio with a prior to get posterior odds is S6 §5; §4 there runs this exact 1:50 calculation.

The threshold is a Bayesian statement, not a p-value. Morton's conventional cutoff Z_{\max} \ge 3.0 means the data are 1,000× more likely under linkage. But two loci picked at random are rarely linked — with 22 autosome pairs and a ~3,400 cM map, the prior odds are roughly 1:50. So

\text{posterior odds} \approx \frac{1}{50} \times 1000 = 20{:}1 \;\Rightarrow\; \approx 5\%\ \text{false positives}

The famous "3" is a 1,000:1 likelihood ratio deliberately chosen to survive a 50:1 prior against. Symmetrically, Z(\theta) \le -2.0 (100:1 against) excludes linkage at that \theta — and exclusion mapping was historically as useful as detection. For a genome-wide scan the multiple-testing burden pushes the bar up, and how far depends on the design. Lander and Kruglyak's 1995 guidelines put for lod score analysis in fully informative human pedigrees at 3.3, with 1.9 as merely suggestive; for -sharing (affected sib-pair) scans the significant threshold is 3.6, with 2.2 suggestive. The pairs go together — quoting a threshold from one row of that table against the suggestive figure from another describes no study anyone actually ran.

10. Forward: from pedigrees to populations

Everything above tracks recombination through one or two meioses of a known . Run the same process for hundreds of generations in a population and the accounting changes but the physics does not: at nearby loci remain statistically associated because too few crossovers have accumulated between them to randomise their pairing. That residual association is (Ch 29), and it is the reason a genotyping array reading one variant per few kilobases can tag almost all common variation — which is the entire operating premise of (Ch 51).

Linkage disequilibrium is the population-level signature that leaves behind when too few meioses have separated nearby alleles. The two are not the same statement. Linkage is a property of two loci — a distance along a , fixed for the species. LD is a property of allele combinations in a particular population, and so it also encodes that population's history: , and selection can generate strong LD between loci on different chromosomes, and two tightly linked loci sit at complete equilibrium if no ever coupled their alleles. Linkage is necessary for LD to persist, not sufficient for it to exist.

What the change of timescale buys is resolution. Linkage has long range and low resolution; LD has short range and high resolution, because population history has supplied millions of meioses instead of ten.


Common misconceptions

What people thinkWhat's actually true
Linked genes are always inherited togetherOnly probabilistically. sets the rate at which the association breaks; at 20 cM, one gamete in five is recombinant
RF = 50% means the loci are far apart on the same It means the measurement saturated. Unlinked and very-distantly-linked are indistinguishable by RF, which is precisely why maps are chained from short intervals
Map distance and are the same thingEqual only in the small-distance limit. d counts expected crossovers; r counts odd numbers of them. r is bounded by 0.5; d is unbounded
A recombinant gamete carries a new It carries a new combination of existing alleles. Recombination creates no new variants — only does (Ch 16)
"Recombinant" is a property of the gameteIt is defined relative to the parent's . The same A b gamete is recombinant from a coupling parent and parental from a repulsion parent
The rarest class in a three-point cross is a sampling accidentIt is the double-crossover class, and it is the most informative row in the table — it determines gene order
Double crossovers can be ignoredThey are why two-point distances fail to add, and omitting them from both interval counts is the standard arithmetic error
NPD tetrads are just unusually lopsided tetratypesThey have one origin and one only — a four-strand double crossover — so they are second-order rare and diagnostic rather than noisy. Each NPD you observe stands for three double-crossover tetrads you cannot, which is the entire content of the 3 in the Perkins equation
1 cM = 1 MbA genome-wide average that is locally wrong by orders of magnitude. Recombination is concentrated in ~1–2 kb hotspots; centromeric regions are near-silent
are conserved features of the sequenceIn humans and mice they are positioned by , whose binding array evolves extremely fast. Human and chimpanzee hotspots barely overlap
A LOD of 3 means p < 0.001It is a likelihood ratio of 1,000:1, chosen to overcome a ~50:1 prior against linkage. The resulting error rate is nearer 5% — and for a genome scan the bar is higher still

Worked example: a complete three-point testcross

Three markers on Drosophila 3: scarlet (st, bright red eyes), spineless (ss, small bristles), ebony (e, dark body). A female heterozygous for all three is testcrossed to a triple-recessive male, and 1,000 progeny are scored.

(Drosophila males do not undergo at all, so the heterozygous parent in a mapping cross must be the female. The counts below are constructed for clean arithmetic; the answer lands within half a of the published Drosophila map.)

The loci are listed in the columns in an arbitrary order — deducing the true order is part of the exercise. + denotes the wild-type .

   ss     e     st      count
  ────  ────  ────     ──────
   +     +     +          372
   ss    e     st         367
   +     +     st          72
   ss    e     +           69
   ss    +     st          57
   +     e     +           54
   ss    +     +            5
   +     e     st           4
                        ──────
                          1000

Step 1 — parentals. The two largest classes are + + + (372) and ss e st (367), summing to 739. The heterozygous mother was therefore in coupling: one chromosome carried all three mutant alleles, the other all three wild-type alleles.

Step 2 — double crossovers. The two smallest classes are ss + + (5) and + e st (4), summing to 9.

Step 3 — gene order. Compare a DCO class to the parental class it most resembles:

   parental   +   +   +
   DCO        ss  +   +
              ↑
              only the ss locus has changed

(The reciprocal pair agrees: parental ss e st versus DCO + e st also differs only at ss.) A double crossover flips the middle marker and leaves the flanks alone, so ss is in the middle and the order is st – ss – e. Call st–ss region I and ss–e region II.

Step 4 — assign every class. Rewrite each in map order:

Class (in map order st ss e)CountCrossover in
+ + +372— (parental)
st ss e367— (parental)
st + +72region I
+ ss e69region I
st ss +57region II
+ + e54region II
+ ss +5I and II
st + e4I and II

Single crossovers in region I total 141; in region II, 111; doubles, 9.

Step 5 — interval distances. Double crossovers experienced an exchange in both regions, so they count in both:

\text{RF}_{st\text{–}ss} = \frac{141 + 9}{1000} = 0.150 \;\Rightarrow\; \textbf{15.0 cM}
\text{RF}_{ss\text{–}e} = \frac{111 + 9}{1000} = 0.120 \;\Rightarrow\; \textbf{12.0 cM}

giving the map

   st ──────────── 15.0 cM ──────────── ss ────────── 12.0 cM ────────── e
   |←──────────────────────── 27.0 cM ───────────────────────────────→|

Step 6 — what a two-point cross would have said. Score st against e alone. The DCO classes + ss + and st + e are parental for that pair — the middle marker is invisible in a two-point cross:

\text{RF}_{st\text{–}e} = \frac{72 + 69 + 57 + 54}{1000} = 0.252 \;\Rightarrow\; 25.2\%

against the additive 27.0 cM. The shortfall is exactly 2 \times \text{DCO}/N = 1.8\%: each double crossover costs two recombinants that should have been counted. This is why maps are additive sums of short intervals rather than long-range measurements.

Step 7 — .

\text{expected DCO} = 0.150 \times 0.120 \times 1000 = 18 \qquad \text{observed} = 9
\text{c.o.c.} = \frac{9}{18} = 0.5, \qquad I = 1 - 0.5 = \textbf{0.5}

Half the double crossovers expected under independence were suppressed.

Step 8 — mapping functions on the two-point estimate. Suppose all we had was r_{st\text{–}e} = 0.252. What distance does each function recover?

\text{Haldane: } d = -\tfrac{1}{2}\ln(1 - 0.504) = -\tfrac{1}{2}\ln(0.496) = 0.351 \text{ M} = 35.1 \text{ cM}
\text{Kosambi: } d = \tfrac{1}{4}\ln\frac{1.504}{0.496} = \tfrac{1}{4}\ln(3.032) = 0.277 \text{ M} = 27.7 \text{ cM}

The truth from the three-point data is 27.0 cM. Haldane overshoots by 30% because it assumes no interference and therefore inflates the number of hidden double crossovers; Kosambi, which assumes interference of roughly the magnitude actually present, lands within 0.7 cM. That gap is the practical case for Kosambi, and a reminder that a mapping function is a model with an assumption you can check.

Connections