Backcrossing in Cannabis Breeding – Recurrent Parent, Linkage Drag and the Limits of the Method

Moving one trait from an outside line into an existing variety without losing the variety: that is what the method is for, and the arithmetic behind it is simple. The two conditions under which that arithmetic holds are missing from most guides, and in cannabis it is the second one that tends to fall away.

What the arithmetic promises

Each backcross to the same parent halves the donor’s share. Without any selection, the expected proportion of the recurrent parent genome is 75.0 % after the first round, 87.5 after the second, 93.8 after the third, 96.9 after the fourth and 98.4 % after the fifth. Transferring a single dominant gene by the classical method accordingly takes five or more generations to recover 99 % of the original genome.

Markers in the background bring that number down. Screening every generation not only for the target gene but also for how much donor genome each individual still carries gets the same result in three generations instead of six. That is the real payoff of marker-assisted backcrossing, and it is a gain in time, not in quality.

German breeding literature calls the method Verdrängungszüchtung, displacement breeding. It belongs where a monogenic trait – the resistance of a landrace, say – is to be combined with the trait complex of a high-yielding but susceptible variety.

The two conditions rarely stated

First, the recurrent parent should itself be a true-breeding line. If it is not, it segregates along with everything else in every round, and what returns is not the variety but a sample of it. In cannabis that is the rule rather than the exception: an open-pollinated accession carries between 17 and 35 % heterozygous loci. A true-breeding recurrent parent therefore has to be built first, by sib-mating or by chemically induced selfing, and either costs generations before the first backcross begins.

Second, the trait being moved should rest on one or a few loci. For a monogenic trait, selection in each round is unambiguous; once several loci are involved, the chance of finding them all in the same individual drops with every added gene. Marker-assisted backcrossing is therefore most effective with one or a few target genes and is the wrong tool for a quantitative trait.

Linkage drag

What does not thin out along with the rest is the immediate neighbourhood of the target gene. Because every round selects for that one allele, the piece of donor genome attached to it survives each dilution: the linkage drag.

How large that piece stays depends not on the number of generations but on how often the region recombines at all. In a simulation on oilseed rape the dragged segment averaged 15.9 million base pairs after the first backcross and 5.4 million after the second; after three rounds followed by selfing, 98 % of the remaining donor genome came from that single segment. Where the target gene sat in a region of brisk exchange, the same piece shrank to a fraction of a megabase.

Why this weighs more heavily in cannabis

The genome structure of this species works against the method. Across 78 haplotype-resolved genomes, linkage disequilibrium decays to half its maximum only after roughly 10 kilobases, comparable to wild outcrossing soybean and rice populations, and individual marker pairs stay linked across hundreds of kilobases into the megabase range.

Then there are the inversions. Each genome carries 86 on average, the mean length is 304 kilobases and the longest reach 25 megabases. Within an inverted stretch, effectively no exchange happens between the two versions – whatever sits inside travels as a block. The clearest case in the surveyed material is a 19.5-megabase inversion on chromosome 1 holding around 1,203 genes, among them the flowering regulator PRR3, which has been linked to day-neutral flowering. Introgressing that trait moves a neighbourhood, not a gene.

On top of that, segregation distortion has been observed in several regions of the cannabis genome: one parent’s alleles come through more often than the arithmetic predicts. For a backcross programme this means the background does not return at the same rate everywhere.

The measured case: the chemotype locus

The best-known introgression in this species turns on exactly this point. The CBD-rich hemp lines that produced, among others, the CBDRx reference genome arose by crossing the CBDAS region into a predominantly drug-type background. The purpose was to pair the high-yield genetics of drug-type plants with a CBD chemotype.

The region in question is a poor candidate for a clean separation. Each haploid genome carries at most one full-length copy of THCAS or CBDAS, embedded in conserved cassettes of transposable elements 70 to 80 kilobases long, usually alongside pseudogene copies of the same family. These cassettes occur in only a few fixed arrangements, sit at different positions between use types and are associated with inversions – yet all of it is packed into roughly 1.5 megabases on chromosome 7. The map from a cross between a drug type and a hemp variety found not one recombination event in the CBDAS block among 99 offspring.

In practice: whoever crosses in the chemotype crosses in the cassette, and selection cannot take it apart. That chemotype segregation still follows the simple single-locus model has the same cause and is no sign of a simply built trait.

What the percentage does not measure

The share of recovered parental genome is normally scored on single-nucleotide markers, and in this species their heterozygosity runs between 1 and 2.5 %. Counting structural variants and stretches that cannot be aligned between two haplotypes at all, an average of 20.6 % of genome length is variable. A background that looks 98 % recovered by marker score can differ in precisely the regions no marker covers.

A second limit concerns the choice of donor. The fixation index between drug type and hemp stands at 0.20, a value otherwise seen between species. A cross across that distance may not merely fail to raise vigour but lower it. How far the donor may sit from the recurrent parent is therefore not an academic question.