Linkage, crossing over, and genetic maps: LOD scores and recombination frequencies

Intuition Beginner

Mendel's law of independent assortment holds for genes on different chromosomes, but genes that sit close together on the same chromosome tend to be inherited as a package — this is called linkage. During meiosis, homologous chromosomes can swap segments through crossing over, which separates linked genes. The farther apart two genes are on a chromosome, the more often a crossover will land between them. By counting how often genes are reshuffled, scientists build genetic maps that show the order and spacing of genes along chromosomes.

Visual Beginner

Worked example Beginner

In Drosophila, genes for body colour (grey B, black b) and wing shape (normal V, vestigial v) are linked on the same chromosome. A dihybrid fly in coupling phase (BV/bv) is test-crossed to bv/bv. Among 1000 offspring: 418 are BV/bv, 402 are bv/bv, 90 are Bv/bv, and 90 are bV/bv.

The recombinant types are Bv/bv and bV/bv (non-parental combinations). Recombination frequency = (90 + 90)/1000 = 0.18, or 18%. This means the two genes are approximately 18 map units (centiMorgans) apart.

Check your understanding Beginner

Formal definition Intermediate+

Complete and incomplete linkage

Complete linkage occurs when two loci are so close on the same chromosome that no recombination is observed between them. All gametes from a dihybrid are parental types.

Incomplete linkage is the general case: some gametes are parental (non-recombinant) and some are recombinant. The recombination frequency $r$ satisfies $0<r<0.5$.

Recombination frequency

For a test cross of a dihybrid in coupling phase ($AB/ab$) crossed with $ab/ab$:

$$r=\frac{\text{number of recombinant offspring}}{\text{total offspring}}$$

The recombination frequency $r$ ranges from 0 (complete linkage) to 0.5 (independent assortment, either on different chromosomes or far apart on the same chromosome).

Map units and centiMorgans

One centiMorgan (cM) equals 1% recombination frequency. A distance of $d$ cM corresponds to $r=d/100$ for small $r$ (roughly $r<0.2$). For larger distances, the relationship between $r$ and true map distance is nonlinear because multiple crossovers between two loci can restore the parental configuration, causing $r$ to underestimate the true number of crossovers.

Two-point and three-point crosses

A two-point cross measures recombination between two loci simultaneously. It provides a direct estimate of $r$ between those loci.

A three-point cross involves three linked loci (e.g., $A$, $B$, $C$) and is more informative. It simultaneously estimates two recombination frequencies ($r_{AB}$ and $r_{BC}$), determines the gene order (which gene is in the middle), and detects double crossovers (simultaneous recombination in both intervals).

Gene order determination: the rarest classes of offspring correspond to double crossovers. The allele that is swapped relative to the other two in the double-crossover classes identifies the middle gene.

Interference and coefficient of coincidence

Interference ($I$) measures the reduction in double crossovers relative to expectation under independence:

$$I=1-C$$

where $C$ is the coefficient of coincidence:

$$C=\frac{\text{observed double crossovers}}{\text{expected double crossovers}}$$

Expected double crossovers = $r_{AB}\times r_{BC}$. Positive interference ($I>0$, $C<1$) means fewer double crossovers than expected. Negative interference ($I<0$, $C>1$) means more.

Mapping functions

The Haldane mapping function assumes no interference (crossovers occur as a Poisson process along the chromosome):

$$d=-\frac{1}{2}\ln (1-2r),r=\frac{1}{2}(1-e^{-2d})$$

where $d$ is the map distance in Morgans (1 M = 100 cM). This corrects for unobserved double crossovers.

The Kosambi mapping function accounts for interference:

$$d=\frac{1}{4}\ln \left(\frac{1+2r}{1-2r}\right),r=\frac{1}{2}\left(\frac{e^{4d}-1}{e^{4d}+1}\right)$$

The Kosambi function is preferred for organisms with moderate interference, because it models the observed reduction in double-crossover frequency.

Tetrad analysis

In some fungi (e.g., Neurospora, Sordaria), all four products of a single meiosis can be recovered as an ordered tetrad. This allows direct observation of whether a crossover occurred, rather than inferring it from offspring phenotypes. Tetrads are classified as:

• Parental ditype (PD): only two spore types, both parental.
• Non-parental ditype (NPD): only two spore types, both recombinant.
• Tetratype (T): four spore types (two parental, two recombinant).

The frequency of NPD tetrads distinguishes linkage from independent assortment: for unlinked loci, PD $\approx$ NPD; for linked loci, PD $\gg$ NPD.

LOD scores

The LOD score (logarithm of odds) tests whether two loci are linked:

$$Z(\theta )={\log }_{10}\left(\frac{L(\text{data}\mid \text{linkage at recombination fraction }\theta )}{L(\text{data}\mid \text{independent assortment, }r=0.5)}\right)$$

A LOD score of $Z=3$ (odds of 1000:1 in favour of linkage) is the conventional threshold for declaring linkage. A LOD score of $Z=-2$ is the threshold for rejecting linkage. The recombination fraction $\hat{\theta }$ that maximises $Z(\theta )$ is the maximum-likelihood estimate of the map distance.

For $n$ informative meioses with $k$ recombinants:

$$Z(\theta )={\log }_{10}\left(\frac{{\theta }^k(1-\theta {)}^{n-k}}{(0.5{)}^n}\right)$$

Key result Intermediate+

Result (Haldane mapping function). Under the assumption that crossovers occur as a Poisson process with no interference, the relationship between recombination frequency $r$ and map distance $d$ (in Morgans) is $d=-\frac{1}{2}\ln (1-2r)$.

Derivation. Let $X$ be the number of crossovers in an interval of length $d$. Under the Poisson model, $X\sim \text{Poisson}(2d)$ (the factor 2 arises because each crossover involves two of the four chromatids, so on average half of the chromatids are affected). An odd number of crossovers produces a recombinant chromatid; an even number restores the parental configuration. The probability of recombination is:

$$r=P(X\text{ odd})=\sum _{k=0}^{\infty }\frac{e^{-2d}(2d{)}^{2k+1}}{(2k+1)!}=\frac{1}{2}(1-e^{-2d})$$

Solving for $d$: $1-2r=e^{-2d}$, so $d=-\frac{1}{2}\ln (1-2r)$. Note that as $d\to \infty$, $r\to 1/2$, reflecting the ceiling effect where multiple crossovers obscure the true distance. $\square$

Result (LOD score threshold). The conventional threshold LOD $\ge 3$ for declaring linkage corresponds to a likelihood ratio of 1000:1 in favour of linkage over independent assortment. This threshold accounts for the prior probability that two random loci are linked (approximately 1/22 for human autosomes, since there are 22 autosomal linkage groups among 23 chromosome pairs), providing a posterior probability of linkage above 0.95 when the threshold is met.

Result (Three-point cross gene order). Given three linked loci $A$, $B$, $C$ and offspring from a test cross of a trihybrid, the gene in the middle is identified by finding the allele that switches relative to the other two between the single-crossover and parental classes. The double-crossover classes (the rarest) directly reveal the middle gene.

Exercises Intermediate+

Advanced topics in linkage analysis Master

Multipoint linkage analysis

Multipoint linkage analysis extends the two-point LOD score to multiple markers simultaneously. Given marker data at loci $M_1,M_2,\dots ,M_k$ and a putative disease locus $D$, the likelihood is computed over all possible phase-known haplotype configurations:

$$L({\theta }_1,{\theta }_2,\dots ,{\theta }_k)=\sum _{\text{haplotypes}}\prod _{\text{meioses}}P(\text{offspring haplotype}\mid \text{parental haplotypes},{\theta }_1,\dots ,{\theta }_k)$$

The Elston-Stewart algorithm (1971) performs this computation efficiently for pedigrees by peeling: summing over unobserved genotypes one nuclear family at a time. The Lander-Green algorithm (1987) uses hidden Markov models along the chromosome, treating the inheritance vector (the pattern of grandparental origin at each locus) as a hidden state. The Elston-Stewart algorithm scales well with the number of markers but poorly with pedigree size; the Lander-Green algorithm scales well with pedigree size but poorly with the number of markers (the state space is $4^n$ where $n$ is the number of non-founders).

Haplotype blocks and linkage disequilibrium

A haplotype is the specific combination of alleles at multiple loci on a single chromosome. In populations, certain haplotypes occur more frequently than expected under independent assortment — this is linkage disequilibrium (LD). The two standard measures are:

$$D=P_{AB}-p_A{p}_B,D^{\prime }=\frac{D}{D_{\max }}$$ $$r^2=\frac{D^2}{p_A{p}_a{p}_B{p}_b}$$

where $P_{AB}$ is the observed frequency of haplotype AB and $p_A,p_B$ are marginal allele frequencies. $D^{\prime }$ ranges from $-1$ to $1$ and measures the departure from linkage equilibrium; $r^2$ ranges from $0$ to $1$ and measures the predictive power of one marker for another.

Haplotype blocks are chromosomal regions within which only a few common haplotypes are observed, separated by recombination hotspots. The HapMap project (2002-2009) catalogued LD patterns across the human genome and found that about 70% of the genome falls within haplotype blocks where a small number of tagging SNPs capture most of the common variation. This structure is exploited in genome-wide association studies (GWAS): genotyping a subset of tagging SNPs is sufficient to impute the remaining variants.

LD decays over time due to recombination at rate approximately $D_t\approx D_0(1-r{)}^t$ per generation. In large outbred populations (e.g., Europeans), LD typically extends 10-30 kb; in bottlenecked or recently founded populations (e.g., Finns, Sardinians), LD extends further (50-100 kb) because fewer generations of recombination have occurred since the bottleneck.

Genetic maps versus physical maps

A genetic map measures distances in recombination units (cM). A physical map measures distances in base pairs (bp). The two are not linearly related because recombination rate varies along chromosomes:

• Recombination is suppressed near centromeres and telomeres.
• Recombination hotspots (1-2 kb regions with greatly elevated crossover rates) account for a disproportionate fraction of total recombination.
• The genome-wide average in humans is approximately 1 cM per 1 Mb, but local rates vary from near zero to over 10 cM/Mb.

The first human genetic map (Donis-Keller et al. 1987) used 403 polymorphic markers (RFLPs) at an average spacing of about 10 cM. Modern maps (deCODE, HapMap, 1000 Genomes) use millions of SNP markers and achieve sub-kilobase resolution. The deCODE map (Kong et al. 2002, 2010) was constructed from 146 families with 1,257 meioses in Iceland and revealed fine-scale recombination rate variation.

Linkage disequilibrium mapping

LD mapping (also called association mapping) exploits the fact that a disease-causing mutation arose on a specific haplotype background. Nearby markers in high LD with the causal variant will show association with the disease. This is the basis of GWAS.

The resolution of LD mapping depends on the extent of LD: in populations with short LD blocks (e.g., Africans), association signals are more localised (higher resolution) but require denser marker sets. In populations with long LD blocks (e.g., Europeans), association signals are broader (lower resolution) but fewer markers are needed.

Fine mapping narrows the associated region by sequencing the region in cases and controls, identifying the specific causal variant among correlated markers. This is nontrivial because multiple variants in high LD may all show statistical association; functional validation (e.g., CRISPR editing, reporter assays) is needed to identify the causal variant.

Morgan and Sturtevant's Drosophila work

Thomas Hunt Morgan discovered sex linkage in Drosophila melanogaster in 1910 through the white-eye mutation ($w$). His student Alfred Sturtevant, at age 21, realised that recombination frequencies could be used to construct linear maps of gene order. In his 1913 paper "The linear arrangement of six sex-linked factors in Drosophila" ^{[Sturtevant 1913]}, Sturtevant reported recombination frequencies among six X-linked genes ($w$, $v$, $m$, $r$, $p$, $s$) and showed that the distances were additive: if $d_{AB}+d_{BC}=d_{AC}$ (approximately), gene B lies between A and C. This was the first genetic map and established that genes are arranged linearly on chromosomes.

The discovery resolved a key question: genes are not abstract entities but have physical positions on chromosomes. The additivity of map distances (with corrections for double crossovers) confirmed the linear arrangement and provided the conceptual foundation for all subsequent genetic mapping.

Modern linkage mapping

SNP arrays (e.g., Illumina Infinium, Affymetrix GeneChip) genotype hundreds of thousands to millions of single-nucleotide polymorphisms simultaneously. In family-based linkage studies, SNP arrays provide dense marker coverage that increases the informativeness of each meiosis: virtually every crossover can be detected because adjacent SNPs bracket the crossover point.

Microsatellites (short tandem repeats, STRs) were the standard markers for human linkage mapping before SNP arrays. Their high polymorphism (5-20 alleles per locus) makes them highly informative for linkage: the probability that a meiosis is informative (heterozygous parent) is high. The first human linkage maps (Genethon, 1996; deCODE, 2002) were constructed primarily from microsatellites.

Nonparametric linkage methods do not require specifying a genetic model (dominant/recessive, penetrance). The allele-sharing method (affected sib-pair test) compares the observed number of alleles shared identical-by-descent (IBD) among affected relatives to the expectation under no linkage. For sib pairs, the expected IBD sharing under no linkage is 1/2 for each allele; excess sharing at a marker indicates linkage to a disease locus.

Recombination hotspots and PRDM9

In humans, most recombination occurs in narrow (1-2 kb) hotspots that collectively account for 60-80% of all crossovers. The gene PRDM9 (PR domain-containing protein 9) is the major determinant of hotspot location in mammals. PRDM9 encodes a zinc-finger protein that binds specific DNA sequence motifs and deposits the H3K4me3 histone mark, initiating the double-strand breaks that lead to crossovers.

Different alleles of PRDM9 recognise different sequence motifs, causing hotspot locations to differ between individuals and populations. PRDM9 evolves rapidly under positive selection, and the hotspots it creates are self-destructing: the gene conversion bias that accompanies recombination erodes the PRDM9-binding motif, causing hotspots to decay over evolutionary time. This "hotspot erosion" explains why PRDM9 must continually evolve new specificities.

In organisms lacking PRDM9 (e.g., Saccharomyces cerevisiae, Arabidopsis thaliana), recombination hotspots coincide with promoter regions and other features of open chromatin rather than with specific sequence motifs.

Sex-specific genetic maps

Recombination rates differ between the sexes. In humans, the female genetic map is approximately 1.6 times longer than the male map (about 4400 cM vs 2700 cM), reflecting higher overall recombination in females. The sex difference varies along chromosomes: near telomeres, male recombination approaches or exceeds female recombination, while in interstitial regions female recombination is substantially higher.

Sex-specific maps are important for linkage analysis because the sex of the informative parent determines which map to use. In X-chromosome mapping, only female meioses provide recombination information (male meioses involve no X-X recombination).

Gene conversion

Gene conversion is a nonreciprocal transfer of genetic information from one homologous sequence to another during meiosis. Unlike crossing over, gene conversion does not exchange flanking markers — it alters the sequence at a single site within a short tract (typically 50-2000 bp).

Gene conversion matters for linkage analysis because it can create apparent recombinants between very closely linked markers without an accompanying crossover. In fungi, where all four meiotic products are recoverable, gene conversion events appear as 3:1 or 4:0 segregation ratios at a single locus within an otherwise non-recombinant tetrad. The frequency of gene conversion tracts (estimated at $5\times 10^{-5}$ to $5\times 10^{-4}$ per locus per meiosis in yeast) complicates fine-structure mapping at the sub-kilobase scale.

Result (Additivity of map distances). For three loci $A$, $B$, $C$ in linear order with map distances $d_{AB}$ and $d_{BC}$, the map distance $d_{AC}=d_{AB}+d_{BC}$ holds exactly under the Haldane model (no interference). Under interference, the observed recombination frequency for the combined interval is less than $r_{AB}+r_{BC}-2r_{AB}{r}_{BC}$ (the expectation without interference).

Result (LOD score for a phase-known pedigree). For a phase-known pedigree with $n$ informative meioses and $k$ recombinants at recombination fraction $\theta$, the LOD score is $Z(\theta )=k{\log }_{10}(2\theta )+(n-k){\log }_{10}(2(1-\theta ))$. The maximum is attained at $\hat{\theta }=k/n$.

Connections Master

• Mendelian genetics 19.01.01. Linkage is the exception to Mendel's law of independent assortment. Genes on the same chromosome do not assort independently; recombination frequency quantifies the departure from independence and provides a natural metric for gene order and distance.

• Population genetics 19.02.01. Recombination generates the haplotype diversity that population genetics models describe. Linkage disequilibrium between loci decays at rate $(1-r{)}^t$ per generation, connecting recombination frequency to the rate at which population haplotype structure equilibrates.

• Quantitative genetics 19.05.01. Linkage between a quantitative trait locus (QTL) and a molecular marker enables QTL mapping: the recombination fraction between marker and QTL determines the resolution with which the QTL can be localised. Multipoint linkage analysis provides the statistical framework for interval mapping.

• Phylogenetics 19.07.01. Recombination disrupts the tree-like structure of genealogical relationships at linked loci. In regions of low recombination, the genealogy of a chromosomal segment approximates a single tree; in regions of high recombination, different sites have different trees. This mosaic genealogy is modelled by the coalescent with recombination.

• Molecular biology of meiosis 17.08.01. Crossing over is initiated by programmed double-strand breaks catalysed by Spo11, processed by the MRN complex, and resolved as either crossovers or non-crossovers (gene conversions) by the ZMM pathway or the Mus81-Mms4 resolvase. The molecular machinery determines the distribution and frequency of recombination events that genetic maps measure.

Historical notes Master

The concept of genetic linkage emerged from Morgan's Drosophila laboratory at Columbia University between 1910 and 1916. Morgan's 1910 discovery of the white-eye mutation ($w$) on the X chromosome established sex linkage and demonstrated that genes reside on specific chromosomes. He and his students observed that certain pairs of genes did not assort independently — they tended to be inherited together — and interpreted this as physical linkage on the same chromosome.

Alfred Sturtevant, an undergraduate in Morgan's lab, produced the first genetic map in 1913 ^{[Sturtevant 1913]}. He reasoned that if genes are arranged linearly on chromosomes, and if recombination frequency increases with physical distance, then recombination frequencies should be approximately additive. By measuring pairwise recombination frequencies among six X-linked genes, he constructed a consistent linear map — a result so elegant that it convinced the remaining sceptics of the chromosome theory of inheritance.

Calvin Bridges demonstrated non-disjunction in 1916, providing cytological proof that genes are carried on chromosomes. Herman Muller showed in 1916 that interference (the reduction of double crossovers near existing crossovers) is a general phenomenon, ruling out the Poisson model for crossover distribution along chromosomes.

J.B.S. Haldane introduced the mapping function bearing his name in 1919, providing the mathematical framework to convert recombination frequencies into corrected map distances. D.D. Kosambi introduced the mapping function accounting for interference in 1944.

The LOD score method was developed by Newton Morton in 1955 ^{[Ott 1999]}, building on the sequential probability ratio test of Wald (1945). Morton proposed the threshold of LOD $\ge 3$ for declaring linkage, which became the standard in human genetics. The method was transformative because it allowed linkage information from multiple families to be combined additively (LOD scores from independent families are summed), enabling the detection of linkage for rare Mendelian disorders.

The first human gene mapped by linkage was the Duffy blood group ($FY$) to chromosome 1 by Donahue et al. in 1968, using a family in which a variant chromosome 1 (visible as an enlarged centromere) cosegregated with the Duffy blood type. The human genome project ultimately produced a complete genetic map with over 10,000 markers, enabling the positional cloning of hundreds of disease genes including CFTR (cystic fibrosis, 1989), BRCA1 (breast cancer, 1994), and HTT (Huntington disease, 1993).

Bibliography Master

@article{Sturtevant1913,
  author = {Sturtevant, A. H.},
  title = {The linear arrangement of six sex-linked factors in {Drosophila}},
  journal = {J. Exp. Zool.},
  volume = {14},
  pages = {43--59},
  year = {1913},
}

@article{Haldane1919,
  author = {Haldane, J. B. S.},
  title = {The combination of linkage values, and the calculation of distances between the loci of linked factors},
  journal = {J. Genet.},
  volume = {8},
  pages = {299--309},
  year = {1919},
}

@article{Kosambi1944,
  author = {Kosambi, D. D.},
  title = {The estimation of map distances from recombination values},
  journal = {Ann. Eugen.},
  volume = {12},
  pages = {172--175},
  year = {1944},
}

@article{Morton1955,
  author = {Morton, N. E.},
  title = {Sequential tests for the detection of linkage},
  journal = {Am. J. Hum. Genet.},
  volume = {7},
  pages = {277--318},
  year = {1955},
}

@book{Ott1999,
  author = {Ott, J\"{u}rg},
  title = {Analysis of Human Genetic Linkage},
  edition = {3rd},
  publisher = {Johns Hopkins University Press},
  year = {1999},
}

@book{HartlClark2007,
  author = {Hartl, Daniel L. and Clark, Andrew G.},
  title = {Principles of Population Genetics},
  edition = {4th},
  publisher = {Sinauer Associates},
  year = {2007},
}

@book{FutuymaKirkpatrick2017,
  author = {Futuyma, Douglas J. and Kirkpatrick, Mark},
  title = {Evolution},
  edition = {4th},
  publisher = {Sinauer Associates},
  year = {2017},
}

@article{Kong2010,
  author = {Kong, Augustine and Thorleifsson, Gudmar and Gudbjartsson, Daniel F. and Masson, Gisli and Sigurdsson, Aslaug and others},
  title = {Fine scale recombination rate differences between sexes, populations and individuals},
  journal = {Nature},
  volume = {467},
  pages = {1099--1103},
  year = {2010},
}

@article{Baudat2010,
  author = {Baudat, Fr\'{e}d\'{e}ric and Imai, Yukiko and de Massy, Bernard},
  title = {Regulation of meiotic recombination via the formation of {PRDM9}-dependent and -independent double-strand breaks},
  journal = {Nature},
  volume = {467},
  pages = {1002--1005},
  year = {2010},
}