This next paper is by Jeff Ross-Ibarra (@jrossibarra) on his paper (along with coauthors) Gerke et al The genomic impacts of drift and selection for hybrid performance in maize arXived here.

Iowa recurrent selection as an evolutionary experiment in hybrid vigor

Maize is an outcrossing species, and was cultivated as such up through the first quarter of the 20th century. Starting in the 1920’s, however, breeders began to abandon open-pollinated maize in favor of hybrid varieties resulting from crosses between inbred lines. Hybrids are often more robust and higher yielding than either inbred parent, a phenomenon known as hybrid vigor or heterosis.

Breeding for hybrid varieties – and presumably increased heterosis – has had a profound impact on diversity across the maize genome. There are at least two important differences from previous breeding efforts: first, breeders select on and work with inbred maize lines rather than mass selection on open-pollinated populations. This results in much smaller effective population sizes, and has implications for recessive traits and deleterious alleles that could be masked in heterozygotes. The second difference is that instead of selecting the best plants per se, breeders now select for inbreds that make high-yielding hybrids. This means a breeder might favor an inbred that itself is not high-yielding if it consistently makes good hybrids when paired with other inbreds.

We set out to study the effects of these breeding strategies on patterns of diversity across the maize genome. We took addvantage of one of the longest-running ongoing experiments on selection for hybrid performance, started in the late 1940’s by the US Dept. of Agriculture’s Agricutural Research Service. Two small (12 and 16) sets of founder inbred lines were randomly mated to create two base populations: the Iowa Stiff Stalk Synthetic (BSSS) and the Iowa Corn Borer Synthetic No. 1 (BSCB1). In addition to its role as an important selection experiment, multiple maize breeding lines have come out of the BSSS population, including the line used for the maize reference genome.

Diversity in the BSSS and BSCB1 is patterned predominantly by drift

Over the course of the experiment we studied, the two base populations underwent 16 cycles of recurrent selection, in which lines from each population were crossed to each other and evaluated for both hybrid and per-se performance. Selected lines were intermated within each population to form the next generation. To investigate the genomic impact of this selection scheme, we genotyped progenitor lines and over 600 individuals from multiple selection cycles using the Illumina MaizeSNP50 SNP array. And because we know the exact crossing and selection scheme used, we can compare the observed changes in genome-wide diversity with strictly neutral crossing simulations using the genotypes of the starting populations.

Both populations steadily lost genetic diversity as they became more diverged from one another, but diversity and divergence between BSSS and BSCB1 can be largely reproduced by simulation without any selection. In fact, principal component analysis clearly reveals changes in population structure and diversity that mirror alterations in rates of inbreeding and effective population size that occurred over the course of the experiment. This indicates the structure is not necessarily related to the phenotypic improvement, but might be a by-product of the breeding scheme. Similar population structure is reflected in a recent broad comparison of US maize germplasm and suggests that much of the diversity and structure of modern maize germplasm has been effected by genetic drift.

Selection efficacy and fixation at regions of low-recombination.

But genetic drift can’t be the whole story in these populations. Numerous experiments have shown that the later populations are superior to their progenitors in terms of hybrid yield and traits important to increased planting density (more plants per acre = more yield). These same trends are observed across North American maize as a whole, suggesting common themes in how maize has improved over time. Selection is difficult to detect in the face of strong genetic drift, especially when the selection has been on traits with complex genetic architectures. However our simulations do detect regions of low heterozygosity in each population that are longer than expected given their genetic distance.

The most striking pattern of these regions is their lack overlap between the two populations. In simple cases, classic overdominance models of heterosis predict that at a single locus, two distinct alleles confer heterozygote advantage when combined. In this case, selection should lead to decreased heterozygosity at a locus in both populations as complementary alleles rise in frequency. We don’t observe this, and neither did a different study that used other populations.

A popular alternative to the over-dominance model is the dominance model, which predicts that heterosis is caused by the complementation of linked recessive deleterious alleles. In this case, multiple haplotypes in the other population may complement a fixed region if most deleterious alleles in maize are rare. Evidence from numerous studies supports a dominance model of heterosis, including findings of excess residual heterozygosity in low recombination regions of a maize mapping population. In regions of low recombination, heterozygosity (and thus complementation) becomes important due to an inabilty to efficiently select for new recombinants in these regions, especially with low effective population sizes. And because of low rates of recombination, a small genetic interval in these regions becomes massive in physical space and encompasses the composite effects of many deleterious loci. We observe fixation in these regions in the BSSS and BSCB1 populations. They are short genetically (1-2 centimorgans), but make up very large fractions of the chromosome. We find that in many cases, these regions have been inherited largely intact from the original population founders, indicating that selection for new haplotype combinations in these regions has been ineffective. Large haplotypes in some cases may have fixed early on in the formation of many breeding programs, and the combination of limited exchange between breeding pools and small effective population sizes has provided little opportunity for selective removal of deleterious alleles. Complementation and the inefficiency of selection in these pericentromeric regions, which span a large portion of the physical genome, may thus explain the difference between hybrid and inbred yield and why it has remained fairly constant.