A Century of Corn Selection

See allHide authors and affiliations

Science  04 Feb 2005:
Vol. 307, Issue 5710, pp. 683-684
DOI: 10.1126/science.1105459

Using conventional selection methods, plant and animal breeders have made many beneficial changes to the yields and composition of crops and livestock (1). Yet we know little about the numbers, effects, and mode of action of the genes that account for these long-term changes. A recent paper about maize selection in the journal Genetics demonstrates that such information is slowly becoming available (2).

Since 1896, in one of the longest experiments ever, biologists at the University of Illinois have continuously selected maize (corn) to change the oil composition of its kernels (1, 3). Separate maize lines have been selected for more than 100 generations according to whether the kernels contain high or low amounts of oil, a trait of agronomic importance (3). Typically, mean oil concentration was estimated in 60 or so ears (cobs) of maize, and seeds from only 12 were selected to propagate the next generation. The change in oil concentration was almost continuous (3) and substantial: From a base of about 5%, the high oil-producing line now has about 20% oil in the kernel, and the low oil-producing line has almost none (see fig. S1). The two maize lines differ by about 32 standard deviations (SD, 0.42% in the base population). Divergent selection in separate lines for kernel protein concentration gave similar responses, except that the low line reached a plateau at about 5% protein (3).

To explain the large response in terms of changes at the level of individual gene loci, Laurie, Dudley, and colleagues from the Monsanto Company and the University of Illinois recently reported an analysis of the maize data that itself took much time and work (2). Their goal was to identify quantitative trait loci (QTLs)—regions of the genome where genes influence the trait—by testing the association between markers and the trait, a standard technique of QTL analysis (4).

A cross between high and low oil-producing maize lines from generation 70 was randomly bred for 10 generations in a large population (2) to reduce the effects of linkage disequilibrium. From each of 500 inbred lines subsequently derived by self-pollination, DNA was extracted for genetic analysis and oil concentration was estimated in the ears of inbred plants and of hybrid plants obtained by outcrossing. The single-nucleotide polymorphisms (SNPs) chosen as markers of whether a region of the genome came from either the high or low oil-content line differed substantially in allele frequency between the lines or, exceptionally, filled gaps in the genetic map. After eliminating markers with strong linkage disequilibrium, the investigators focused their analysis on 440 SNPs. Because markers more than 20 cM apart were essentially in linkage equilibrium, real associations would be expected only between close markers and the trait of interest.

An analysis of variance that fitted each marker locus individually and in pairs in separate analyses of inbred and hybrid data showed that both dominance within loci and epistatic interactions between pairs of loci were weak relative to additive effects. This means that the oil concentration in heterozygotes was intermediate between that of the two homozygotes, and that effects at different gene loci did not interact. The high correlation (0.75) of QTL effects on oil content estimated from inbred and hybrid plants is a further indication that they act additively. With the use of a stepwise multiple regression analysis to select markers linked to QTLs and to account for linkage disequilibrium between them, 50 markers were selected for the inbred data and 39 for the hybrids (where differences are smaller). Significant effects were found on all 10 chromosomes, with some clustering in the genome.

A major problem in QTL analyses comprising many tests of significance is to compromise between declaring false associations while missing real ones. To assess their findings, Laurie et al. simulated data with effects distributed similarly to what they observed, and subjected these data to the same analysis. They concluded that they had detected about 63% of the QTLs and that about 33% of markers selected were not QTLs. Consequently, they calculated the correct number of QTLs to be about 50.

The estimated effects of the QTLs on oil concentration were all much less than the line divergence of 15% at generation 70. Indeed, the largest had an effect (half homozygote difference) of about 0.3% oil, and most had a difference of 0.1 to 0.2%. Only about 80% of the QTLs had effects in the same direction as the selection response (38 of 50 inbreds, 33 of 39 hybrids). For the others, the marker allele from the high oil-producing line was associated with low oil—however, alleles of small effect can be fixed by chance in the opposite direction to that of selection if few parents are selected (5). The detected QTLs, if segregating independently, could account for about half the genetic variance of the trait in the population and, by summing their effects, about half the divergence between the high and low oil-producing maize lines. The remaining QTLs are likely to have similar or smaller effects on oil concentration because those with large effects were unlikely to be missed given the degree of genome coverage and the size of the experiment.

Similar findings have also been obtained in a recent study on divergent selection lines of poultry. Andersson and colleagues (6) analyzed a large intercross of poultry lines that had been selected by Siegel in Virginia (7) for high and low body weight (at 8 weeks of age) for 40 generations and that differed in body weight by a factor of about 9. Although 13 QTLs were detected, none individually accounted for more than 3% of the body-weight variance in the F2 generation, and each of these QTLs contributed only a small part of the divergence between the selected lines. Furthermore, the QTLs mainly had additive effects on body weight, as Laurie and colleagues found in their maize analysis.

In most other studies, however, QTLs of substantial effect have been detected— for example, QTLs for body size in poultry, not only in broiler × layer (8) crosses, but also in commercial broiler populations still segregating under intense selection (9). Some QTLs exerting large effects on the trait of interest found in mapping experiments have subsequently been identified as a single causative mutation—such is the case with the mutation in the gene encoding insulin growth factor-2 (IGF-2) in the pig, which alters muscle growth in these animals (10). Although such effects are real, effects of QTLs declared significant tend to be biased upward, and those of small effect are more likely to be missed (4). Models of the underlying distribution of gene effects indicate an exponential form, with numbers increasing as effects get smaller (11). Too much variation is therefore usually attributed to QTLs of large effect.

The recent studies of selected maize and broiler lines (3, 6) were extensive, and the QTL effects identified were small. These appear to conform to the infinitesimal model of genes of small effect assumed in much quantitative genetic theory (4), which predicts the observed continuous steady responses to artificial selection. It is moot as to what defines a “small” effect, however. A maize line containing 0.2% oil in the kernel represents a difference of almost 1 SD between homozygotes in the maize base population; the largest effects detected in the chickens were almost as big (the variance in the F2 was much higher than in the base). The continuing responses to selection, therefore, are not likely to be due mainly to continuing tiny changes in gene frequency predicted by the infinitesimal model; instead they may be due to the fixation of genes, including those arising by mutation after selection started (12, 13), which have appreciable effects while segregating. The biological processes leading to oil concentration or chicken growth are obviously highly interactive, but genes that contribute to selection response must differ in effect when averaged over all other segregating genes. This may explain the Laurie et al. finding that their detected QTLs had approximately additive effects on oil production in maize. We have yet to discover how such QTLs work, but several of the SNPs associated with oil concentration were at candidate loci (2), so there are opportunities to find out. It is a challenge for geneticists to identify the genes and the molecular changes in them that cause these many small but important differences in quantitative traits. It is these small differences that generate variability in populations, providing fuel for change through the action of natural and artificial selection.


  1. 1.
  2. 2.
  3. 3.
  4. 4.
  5. 5.
  6. 6.
  7. 7.
  8. 8.
  9. 9.
  10. 10.
  11. 11.
  12. 12.
  13. 13.
View Abstract

Stay Connected to Science

Navigate This Article