October 9, 1999

Breeding Theory and the Development of Alfalfa

D.E. Rowe and R.R. Hill, Jr.

USDA-ARS, 810 Highway 12 East, Mississippi State, MS 39762 (e-mail derfru@ra.msstate.edu) and USDA-ARS, Curtin Road, University Park, PA 16802 (Deceased), respectively.

ABSTRACT: This is a review of solution oriented research on alfalfa improvement and the development of breeding theory executed by the second author for about 23 years and by the first author for seven years. The researched areas were: (1) selection responses, methods, and their interpretations, (2) development of synthetic varieties, and (3) reality or accuracy of phenotypic measurements. Field and laboratory research could often include multiple cycles of selection, multiple traits selected, or multiple methods of selection. Traits measured and selected by the authors include increased resistances to rust, black stem, pepper spot, stemphylium leafspot, common leafspot, sclerotina crown and stem rots, anthracnose, phytophora root rot, pea aphid, spotted alfalfa aphid, alfalfa blotch miner, and leafhopper. Field performance measurements included yield and height of stands as functions of annual harvest, harvests within years, and age of stand and sward performance as a function of stand densities. Quality components including minerals were also subjected to selection. In a series of experiments spanning 20 years, tests were made in different populations with different traits comparing the gain with selection for the following intra-population improvement procedures: tandem selection, several indexes, phenotypic recurrent selection, polycross family selection, full-sib family selection, half-sib progeny test selection, full-sib family test selection, modified ear-to-row selection, and replicated clonal selection. Inter-populations improvement procedures investigated were reciprocal recurrent selection, test cross to common parent, and the more traditional select and combine procedures. Theoretical expected gains for all of these methods were developed and summarized in a single, terminal monograph. Other research showed the bias and distortions in selection caused by gametic disequilibrium found in the synthetic cultivars and crosses. Several experimental and theoretical studies showed the responses of replacing clones in a small synthetic variety and the effects of increasing the number of parents in the parental or Syn-0 generation. Results suggested that general combining ability (GCA) was the single criteria for selection of best parents for the synthetic cultivar. Practical and theoretical research showed the difficulties of evaluating the GCA accurately. A major cumulative result of this research has been to show that breeding effectiveness on alfalfa has been extremely dependent on the populations and specific conditions of the evaluations and that selection gain is never a certainty.

Introduction

The refereed journal is commonly used by scientists as the best and, likely, the single source for current developments in a particular field of research. But when the area of interest has a narrow focus and few scientists are publishing on a topic, the results, over time, appear piecemeal and disconnected. An advantage of the review article is that the time element can be, in part, ignored. This facilitates the comparison of different publications and the treatment of linkages, interpretations, and some commentary which would not be acceptable in a research publication.

The primary objective of this publication is to develop and enunciate the foci, motivation, and linkages of 23 years of research by the late Dr. R. R. Hill, Jr. and seven years of research by his friend and colleague, Dr. D.E. Rowe. In this review, the senior author has cited only those publications which are, in his very personal opinion, more informative and significant to the continued development of alfalfa. Other publications by the authors are ignored.

The reviewed articles are solution oriented, often included many factors at more than one level, and often are linked or integrated with theoretical studies. In some situations, the experimental results or breeding program motivated theoretical studies in attempts to interpret or predict results in the near and distant future and which, in turn, defined the focus of future plant studies. Both authors regularly modified and applied advanced statistical or genetical analysis to data in an effort to maximize the extraction of information from each study.

The reviewed research has been loosely grouped into three areas: (1) selection methods and their responses in practice and theory, (2) synthetic cultivar development with selection of parents and the measure of their performance, and (3) measurement of phenotype with respect to its accuracy, validity, and interpretation.

Selection Methods and Population Improvement.

Traits used in selection can be loosely classified as either of two types. First are those which are not known to have been naturally exposed to selection pressure and the presence of any genetic variability is unknown. These traits include mineral concentrations, cell structure such as lignin, and the quality or feed value components. The second type of trait includes those for which genetic variability is anticipated because they have obviously impacted the probability of plant survival or number of seed produced. The usual assumptions are that if the species has been exposed to a stress such as that caused by disease or drought at some point in its evolution, then selection pressure has been applied to this host and some genetic variability will be found in the population of survivors. Of course, this does not ensure that genes conferring resistance or tolerance exist or that we have an accurate measurement of the trait. This category of traits includes growth characteristics and resistances to plant pests.

Plant studies made by the authors usually served two purposes: (1) develop improved germplasm which might be released to plant breeders and/or data which could be incorporated into current alfalfa improvement programs and (2) develop information about the genetic or breeding system which has applicability to alfalfa and other autotetraploid crops. Recognition needs to be made of the critically important contribution of expert cooperators who effected the disease screening and measured various plant constituents. Without their cooperative efforts most of the plant breeding research would not have been possible.

The breadth of Dr. Hills research is shown by the number of traits used in the selection studies. The plant diseases screened included rust (Uromyces striatus Schroet.), pepper spot (Leptosphaerulina biosiana Poll.), stemphylium leafspot (Stemphylium botryosum Wallr.) common or Pseudopeziza leafspot (Pseudopeziza medicaginis (Lib.) Sacc.), Sclerotinia crown and stem rot (Sclerotinia trifoliorum Eriks or S. sclerotiorum Lib.), spring black stem (Phoma medicaginis Malbr. & Roum; formerly Asochyta imperfecta), anthracnose (Colletotrichum trifolii Bain & Essary), bacterial wilt (Corynebacterium insidiosum (McCull.) H.L. Jens., bacterial stem blight (Pseudomonas syringae van Hall), Phytophthora root rot (Phytophthora megasperma. Drechs.) Insects investigated were the leafhopper (Empoasca fabae Harris), pea aphid (Acyrthosiphon pisum Harris), spotted alfalfa aphid (Therioaphis maculata Buckton), and alfalfa blotch miner (Agromyza frontella Rondani). Plant components or characteristics measured included all of the usual measures of plant growth such as weight and height for annual performance, individual harvests within a year, and age of stand. Other measurements included concentrations of calcium, phosphorus, potassium, magnesium, sodium, chlorine, copper, zinc, iron, and manganese. Measurements were also made of feed value as estimated by acid detergent fiber, acid detergent lignin, neutral detergent fiber, in vitro dry matter disappearance, and in vitro cell wall digestibility.

Many genetic studies published in quality journals involve one or more cycles of selection on a single population for resistance to a single trait. The results of these programs often lead to what appear to be definitive statements on the presence or absence of gain with selection. The problem then is generalizing these results to the species and to different environments. In contrast, Dr. Hill often executed complex studies using combinations of multiple cycles of selection on multiple traits in different populations while comparing different selection procedures. These complex studies find interactions which are expected to more accurately describe the selection effectiveness but thwart attempts to make simple generalizations about selection procedures or progress. For example, one of Dr. Hill’s research publications from his graduate research reported the results of seven cycles of selection in two populations for resistance to two pests. Measurements were made of changes in genetic parameters for the two selected traits and also for three other traits which had not been subjected to direct selection pressure. Dr. Hill repeated this level of complexity in many genetic studies subsequently executed at University Park, PA.

Though most researchers publish on significant gains with selection, the plant breeder in the private sector may be equally or more concerned with selection that is not effective. Likewise, the private plant breeder is focused on the rate of gain and the biological significance rather than just the statistical significance. For this discussion of selection successes and failures, a summary counting is made of the number of times that a population had significant gain with selection, or significant heritability, or significant genetic variance for seven plant diseases most frequently used in Rowe’s and Hill’s selection studies (Table 1). In this summary, each reported cycle of selection in every population is treated as a separate independent event and given equal weight. One obvious result from Table 1, which is under appreciated by the novice and the uninitiated breeder, is that significance of selection progress or significance of genetic variation is dependent on the population evaluated, the environment, and, apparently, chance. Significant gain with selection or significant genetic variation was highest ( 80%) for the evaluations of rust resistance, but ranged from 12 to 50% for all of the other diseases (Table 1). Certainly one of the most important observations is that significant gain with selection is never a certainty.

An expansion to research on gain with selection for specific traits of alfalfa is research on gain with selection as a function of selection method. Tests of best selection method have involved selecting for resistance to disease using several selection procedures (11) or a combination of procedures and different diseases (5,14), all in the same population(s). Tests have also been made of multiple trait selection methods (2) which is the improvement of a single population for more than one trait with each cycle of selection. The above intra-population improvement procedures were followed by studies on inter-population improvement procedures (22, 23).

The responses to selection were always complex. For instance, four selection procedures, phenotypic recurrent, half-sib family, full-sib family, and alternating selfed and half-sib family, were used in three cycles of selection for one trait in each of two populations (11). The only consistent result was that selection was more effective in one population than the other, but the differences among the selection procedures were not significant. Another comparative test used three different traits (resistance to plant diseases in this case) as a type of replication to compare effectiveness of selection. Selection procedures evaluated were selfed progeny test, selfed family, half-sib progeny test, half-sib family, and phenotypic recurrent (5). With selection for resistance to disease which had relatively high heritability, the selfed family and selfed progeny test selection methods were superior. In contrast, selection for resistance to a disease with lower heritability, showed the phenotypic recurrent selection procedure inferior to the other methods. Another comparative test between half-sib progeny test and the phenotypic recurrent selection used 67 populations (14) with selection on three morphological characteristers and resistances to five diseases. The responses were usually greater using the half-sib progeny test for the three morphological characteristers while the phenotypic recurrent selection was more effective for resistance to the diseases.

Another dimension to tests on intra-population improvement is selection for multiple traits on every individual plant (five diseases and one morphological character) in a single cycle of selection. In a comparison with five cycles of selection using either tandem, independent culling levels, or either of two selection indexes, tandem selection was always inferior to other methods and the indexes were the most effective procedures (2). Tests on inter-population improvement procedures focused on improvement of the cross of pairs of alfalfa populations for three traits and any of three procedures. These results were particularly complex because gain with selection in each population was affected by an unpredictable interaction of gene frequencies of two populations for every trait. The results did not support any general conclusions.

Concurrent with the experimental comparisons were the theoretical studies on expected and potential gain with selection. In diploid species, such theoretical comparisons were completed by Empig et al. (3) in 1972. Theoretical studies on the autotetraploid species began with Hill’s comparisons of diploid and autotetraploids (8). This first effort focused on modeling the genotypes and the development of models for genic action such as additive, monoplex dominance, etc. With adoption of C. C. Li’s model of autotetraploid population effects (16), theoretical comparisons were effected for the following procedures: selfed progeny test, selfed family, half-sib progeny test, half-sib family, full-sib family, and phenotypic recurrent selection (4). This research focused on determination of the expression for the covariance of selection. This covariance relates the frequency of desirable alleles in the selected individuals to the genotypic values of the observed units upon which selection is based. In the case of progeny tests, the selected plants may be the parents of the evaluated progeny populations. In other procedures, the evaluated and selected materials are the same units: individual plants or a family. The equations indicated that for the same selection pressure, the phenotypic recurrent selection should be twice as effective as the half-sib progeny test and the full-sib progeny test and four times as effective as polycross family selection. Expressions for responses with selfing were incompletely developed in this early publication. This research was followed by a expanded and a much more detailed presentation describing the gains with selection for twelve intrapopulation improvement procedures (22). These results are reproduced in Table 2. These expressions include generations per cycle of selection, the covariance of selection, and phenotypic variance for the units upon which selection was based. Gain with selection is maximized when the covariance of selection is large, the number of generations per cycle of selection is minimal, and phenotypic variance is minimal. Different selection methods are directly comparable only when the phenotypic variance (variance among individual plants) is the same for each (Table 2).

Three types of mass selection, labeled 1, 2, and 3, differ with respect to the controls on pollination, but are directly comparable. These procedures were the Mass 1 method or phenotypic recurrent selection where selections are randomly pollinated only by the other selections, the Mass 2 method where selected plants have been pollinated by selected and unselected plants, i.e. the pollen source is not limited to just the selections, and the Mass 3 method where selections are selfed and their selfed progenies are intercrossed to produce the new generation. Mass 1 and Mass 2 procedures require a single generation to complete a cycle of selection, but Mass 3 requires two generations per cycle. Because of the control of pollen sources, the covariance of selection was the same for Mass 1 and Mass 3, but was half as large for the mass 2 selection procedure. Any effects caused by increases in inbreeding were not considered in this comparison. These numbers suggest a comparison of gains should be based on two generations for the particular species. In this time frame, the gain with selection is about the same for Mass 2 and Mass 3, but is only half of that expected for Mass 1 because a cycle of selection is completed in each of the two generations.

Other procedures, which have the same phenotypic variance and are easily compared are some of the progeny tests. These include the half-sib progeny test (HSPT), where parent plants are selected based on performance of half-sib progenies and then intercrossed; the half-sib family selection (HSFS), where the best half-sib progenies are selected and then intercrossed instead of the parental plants; and the half-sib family progeny selection (HSFP), where the members in a half-sib family are intercrossed without crossing between families and the best progenies are selected. The seed from the selected families are planted and crosses are made among the selected families. For the HSFP you have a second generation of crosses but the pollen is only from selections. The covariance of selection of the HSPT, HSFS, and HSFP procedures is in a ratio of 4:2:1, respectively, which approximates the relative expected gains with selection.

Comparisons are also made among the three topcross selection procedures. In the first topcross (TX) procedure, all individuals of a population are pollinated by a representative sample of pollen from a second population and the best parents are selected based on the progeny performances. A second topcross procedure (TXPX) the pollinations and selection criteria are not changed, but the parents have been polycrossed and the seed of the selected parents is used for the next generation. With the third procedure (TXFS) the plants are pollinated as before but the parents are randomly paired as full-sibs and the selections are for pairs of plants based on average performance of the topcross progenies. For all of the topcross methods, the generations per selection cycle is 3 or 4, but the covariance of selection with the TX is twice that of TXPX and TXFS. Thus theoretically, the TX procedure is much more effective.

Other selection procedures included selection of parents based on performance of their selfed progenies followed by intercrossing (S1PT). The covariance of selection for this procedure was simulated for different types of inheritance and it approximated that of the Mass 1 procedure (Rowe, 1981 unpublished). The modified ear-to-row procedure (MER) has two stages: the best half-sib progenies are first selected based on their performance (stage 1) and then the best plants within the selected progenies are selected (stage 2). The final selection procedure was the full-sib family selection (FSF) where random full-sib families are created and selected based on performance of the family progeny and then the families are intercrossed.

These investigations did not attempt to characterize or incorporate the potential effects of inbreeding. Though depression in growth rates, seed vigor, and fertility are often demonstrated in populations exposed to inbreeding, Hill (6) showed that on an individual plant basis this effect is not consistent or predictable and that selection for tolerance to inbreeding is possible. This monograph, by Rowe and Hill (22), was the terminal publication for this research topic.

The theoretical investigations are simple in the assumptions made for their derivation. Imbedded in these derivations is the generally accepted notions that phenotypes have a genetic basis, sometimes defined by the heritability, and that noise or variation in the system which does not reflect genetic differences will reduce the gain with selection. Field predictions of gain are based on estimates of heritability or more simply estimates of the correlation between phenotypic variation and genotypic variation. The authors do have some reservations about the validity of the estimates of the parameter called heritability. The variation among plants or among families and the correlations between parents and offspring are commonly used to estimate the parameter called heritability, but that estimate is postdictive, not predictive. It does not account for any variability in procedures, in environment, or in genetic construct of the plant populations. Thus the usual estimates of h² cannot be incorporated into the predictive equation to generate a meaningful estimate for selection gain without some important modifications which have not been addressed in plant breeding theory.

Though the theoretical studies do not realistically predict gain with selection for any procedure in any specific environment, they do define the relative value of different procedures. They also show that comparisons such as those of HSPT and Mass 1 made on the same populations in Dr. Hill’s experiments (14) are confounded by different variances within and between families, by the number of replicates of each family in the progeny tests, and the number of plants in each replicate. Thus by changing the number of replicates or number of plants in a replicate, the HSPT response changes. Other difficulties in these field tests are differences in selection pressure, sampling errors in the populations, and environmental effects. This ignores that progeny evaluations are based on performance of swards while the mass selections were among spaced plants. The dramatic effects of spacing on performance and on survival were characterized in other field research which showed that these effects and interactions can not be ignored.(17,18).

Synthetic Cultivar Development.

Commercial cultivars are usually synthetic cultivars instead of improved Mass populations because ownership of parents guarantees uniformity of the product and economic protection. Topics of interest to plant breeders lies in the ideal number of parents used to generate the synthetic, the selection of parents for the synthetic and identification of best synthetic cultivar. Unlike germplasm and population improvement research, much of the synthetic cultivar development has been closely linked to theoretical research. Dr. Hill’s earliest theoretical studies (9) focused on defining the linkage between types of genetic effects which are important to synthetic cultivar performance and the population test used to estimate those genetic effects. Probably the most interesting result of Hill’s first study was the observation that specific combining ability (SCA) response was not predictable from a limited number of crosses or with performance in an incomplete mating design. Subsequent research described the relationships between the number of parents used in the synthetic and the probable difference between the mean of the synthetic variety and the mean performance of the original population (7). For maximum gain with selection, one can assume only the best plants of a population are used to generate the synthetic variety. Thus the selected parents will constitute a very small percentage of the total number of plants in a population and in all probability the number of parents used for the synthetic is small. These assumptions are reasonable only if the best parents can be identified and most of the genetic effects of interest are controlled by additive gene action. The implication is that with the greater variance among the plants in a population, the greater the potential gain with selection for the synthetic variety. Thus the plants are accurately ranked and then the best 1, 2, or 5% of the population is used as the parents of the synthetic. The estimates of genetic gain ignore the detrimental effects associated with the inbreeding inherent with using few plants as the parents for the synthetic variety and any negative SCA effects among the parents. Dr. Hill discusses the problems of additive gene action and non-additive gene action on the variance among synthetic cultivars and observes that parents for synthetic cultivars are regularly selected for their SCA responses, but when the synthetic cultivar approaches equilibrium using a small number of selections, most of the crosses will be among closely related genotypes (19).

The phenotypic distance between early generation synthetic evaluation and the performance of the same synthetic cultivar in an advanced generation was further investigated in 1984 (23). This research compared the performance of the Syn₁ generation, which can be treated as a collection of single crosses of the parents, and the performance of that population when it approaches random mating equilibrium. When genic action is not strictly additive, the mean of the Syn₁ generation can be greatly inflated or depressed with respect to that of the advanced generation depending on the gene frequencies involved. Also, the genetic variance of the Syn₁ generation was depressed, particularly when the gene frequency was fixed for one of the parents. Dr. Hill did suggest that selection of the parents for the broad based synthetic should focus on performance of the parents as cloned materials, i.e. their phenotypic performance, not on the performance in crosses. In 1972 Hill et al. published (13) on the measurement of general combining ability (GCA) variance and specific combining ability (SCA) variances with respect to 13 traits in 24 four-clone alfalfa synthetic cultivars originating from one population and 36 four-clone synthetic cultivars from a second population. The rationale of this approach was that selection on basis of GCA or SCA would be effective only if the variation among these synthetic cultivars was significant. Results were very mixed. The variation in GCA values was significant for two diseases for cultivars from both populations and for three diseases in one population. Three of the variances in GCA of morphological traits were significant in both populations, but a fourth was significant in just one population. One assumption is that measurements of GCA are good indicators of parental value in the synthetic cultivar in an advanced generation. The GCA is a statistical measurement to which we ascribe some types of genetic effects, but this does not define a prior the types of genic effects involved. A subsequent theoretical study (19) suggests that for traits controlled by non-additive gene action the crosses involving pollinations with genotypes not in random mating equilibrium, which is the expected case for every synthetic cultivar, can greatly bias the means and thus the estimates of GCA. This bias impacts the probability of errors in selecting the best synthetic cultivars. The theoretical results indicated the GCA estimates may not accurately reflect the value of individual plants as parents for synthetic cultivars. These results also impact the predictive equations developed for estimation of the performance of the advanced generation synthetic from performance of the synthetic in the early generations and the performance of the inbred progenies of the parents.

Phenotypes and inferred linkages to genic effects:

This third topic which has been researched is much more basic for all of plant and animal breeding. The problems with linking or correlating a phenotype to genic factors are not unique to the autotetraploid plant or to alfalfa. Hill’s earliest selection research discussed the impacts of mis-classification in the disease evaluations and the effects of "escapes" on gain with selection (11,12). These escapes were the plants classified as resistant based on a test but were genetically susceptible. Dr. Hill and several other scientists focused their research on allocation or distribution of resources for replicated measurements on same plant or population versus fewer replications and a larger populations (8).

The common concept of phenotype found in textbooks defines the phenotype "P" of a plant or population as a simple function of genetic (G) and environmental (E) effects, their interaction (GxE), and error e such that P = G + E + GxE + e. This mathematical representation is so simple that many, including plant breeders, have an erroneous or incomplete appreciation of the problems in linking phenotype value to genetic effects. The most obvious problem in this algebraic expression is that E is never a constant and that the estimates of G and E and their interaction are based on measurements of P which can not be statistically independent of each other. That is, the estimation of E is always a function of which genotypes are included in the evaluation. Other problems also impact this expression to the extent that interpretations are approximate at best. These topics were the subject of several discussions by the authors, but only rarely were the subject for further research.

The problems with phenotype is elucidated by research on, probably, the most common measurement of phenotype used by plant breeders, the variety trial. The interaction of environment and genotype for these trials was investigated using three statistical approaches in a 12 environment test with 49 entries (10). The analysis of variance in this study showed a significant entry by location by environment interaction and the use of orthogonal contrasts did lend itself to some interpretations on responses over time. The primary observation was that the level of disease resistance in alfalfa was proportional to the performance of the aging stand. The actual component known as GxE was not estimated and, even though it appears logical that disease resistance impacted performance, the linkage was not proven–only suggested. Indeed, the genetic effects contributing to better performance of the aging plot could be only be genetically linked and not physiologically linked to performance. This retrospective interpretation does not constitute verification of cause and effect.

In a later study (15), seven statistical analysis were employed on 14 alfalfa germplasm trials so a total of 150 entries were evaluated for seven years. Though neither of these studies were researched to determine the magnitude of GxE effects, the intent was to determine the true G value so that the entries could be accurately ranked for selection purposes. A rather sophisticated statistical analysis, best linear unbiased prediction method, appeared to be the most efficient method and the expressing of performance as a percentage of check cultivars proved to be the least stable or reproducible procedure. With cross-validation, the different methods predicted different means and had different error rates which were calculated as the square of the difference between of observed and predicted means. Subsequent studies on plant densities (17,18) showed the interaction between stand density and genotype on rates of stand thinning and on the change in yield with aging of the plot. These studies also showed that initial density predisposed the stands to different rates of thinning and interactions of genotype, yield, and density. The question that remains unanswered is what is the true G value for an entry if it is always affected by density, disease resistance, and a multitude of other factors in the environment?

One of the other problems with phenotype is a function of the type of trait which is measured. Many of the morphological traits, such as yield and height, are directly measurable and have a metric which is in some sense natural. In contrast, many of the disease resistances are estimated using an artificial metric. Values which range from 0 to 9 or 1 to 5, for instance, are assigned to diseased phenotypes with the hope that the genetic basis of resistance is monotonically increasing with the improving disease score and that the genetic distance between assigned levels of resistance going from 0 to 1 and 3 to 4, for instance, are equal. The inaccuracy in these assumptions introduces noise into the analysis which can not be removed by any statistical means. An expected effect of this distortion is to alter the tests on the significance of estimates of GCA and SCA. Dr. Hill discussed the problems with escapes in selection of disease resistant alfalfas (11) and attempted some theoretical research (8). Though escapes are logically considered a likely problem in disease evaluations, their impact was not further elucidated until 1985 (20). This study considered systematic errors in estimates of disease resistance and random errors of disease resistance. Results indicated problems with using the so-called check cultivars in the assays. A deficient or excessive inoculation with disease causing propagules could interact with the genic factors effecting resistance, such as dominant or additive genes, and also interact with frequency of these genes in the selected population. Thus for disease resistance, which is the most commonly used variable in genetic studies on alfalfa, the gain with selection is dependent on estimation of the "true" value of a genotype which in turn was a function of inoculation procedures and rates as well as the genetic distance between disease scores. The authors suggest that the historical improvement of disease resistance, yields, and persistence of alfalfa prove that selection in alfalfa is effective and very significant to its utilization. Maybe, the measures of plant morphology and disease resistances upon which genetical studies are conducted should be better viewed as fluid measurements where the genic effects are always nested within the environment.

Summary:

Deficiencies in genetic studies are shown with comparisons of the evaluations made in variety testing and those made for the genetic studies by both authors and their cooperators. The variety tests, which are often executed over many years, attempt to account for differences in aging of stands and other covariates at multiple locations just to estimate yields. Researchers appreciate that conclusions from the variety trials are often nested within the location and within the variable environment. In contrast are the genetic studies which are used to predict gains, to compare different selection procedures, and to make nearly global statements about the genome and the environments The genetic studies are often made with a brief time line and are executed in conditions which are much more artificial in contrast to the field or production tests. For instance, the inconsistencies between some greenhouse disease evaluations and field resistance is appreciated by most plant breeders. The major point proposed by the authors is that field tests of different selection procedures and other related tests are effected with insufficient replication or environmental exposures to be very accurate or useful. The usual test for selection response even when replicated on several different populations is still nested within the procedure and in the particular environment for inoculation as well as evaluation.

The irony is that with the considerable expenditure in field and laboratory genetic tests few results have been conclusive but the theoretical studies which incorporate simplifying, but understood and accepted phenomena, may have produced much more definitive results. Though theoretical research has not predicted the final result of any field or laboratory test, the theory has elucidated the multitude of factors which can impact gain with selection as well as the impact of population gene frequencies, and other phenomena which would not have been considered if only "brute force" selection tests were the basis for determination of best selection method and procedure. For instance, experience had shown that the first generation synthetic cultivar usually performed better than the third generation of that same cultivar, but the theory research indicated that this is always the case when genic effects are not strictly additive. The authors suggest that theoretical research when understood by the plant breeder is critical to the continued and rapid improvement of alfalfa.

References

1. Dudley, J.W., R.R. Hill, Jr., and C.H. Hanson. 1963. Effects of seven cycles of recurrent phenotypic selection on means and genetic variances of several characters in two pools of alfalfa germ plasm. Crop Sci. 3:543-546.
2. Elgin, J.H., Jr., R.R. Hill, Jr., and K.E. Zeiders. 1970. Comparison of four methods of multiple trait selection for five traits in alfalfa. Crop Sci. 10:190-193.
3. Empig, L.T., C.O. Gardner, and W.A. Compton. 1972. Theoretical gains for different population improvement procedures. Univ. Neb. Agr. Exp. Sta. Bul. MP26 rev. 22p.
4. Haag, W.L., and R.R. Hill, Jr. 1974. Comparison of selection methods for autotetraploids. I. Theoretical. Crop Sci. 14:587-590.
5. Haag, W.L., and R.R. Hill, Jr. 1974. Comparison of selection methods for autotetraploids. II. Selection for disease resistance in alfalfa. Crop Sci. 14:591-593.
6. Hill, R.R., Jr. 1975. Parental inbreeding and performance of alfalfa single-crosses. Crop Sci 15:373-375.
7. Hill, R.R., Jr. 1971. Effect of the number of parents on the mean and variance of synthetic varieties. Crop Sci. 11:283-286.
8. Hill, R.R., Jr. 1971. Selection in autotetraploids. Theor. Appl. Genet. 41:181-1886.
9. Hill, R.R., Jr. 1966. Designs to estimate effects of clone substitution in alfalfa synthetics. Crop Sci. 6:471-473.
10. Hill, R.R., Jr., and J.E. Baylor. 1983. Genotype x environment interaction analysis for yield in alfalfa. Crop Sci. 23:811-815.
11. Hill, R.R., Jr., and K.T. Leath. 1979. Comparison of four methods of selection for resistance to Leptosphaerulina briosiana in alfalfa. Can. J. Genet. Cytol. 21:179-186.
12. Hill, R.R., Jr., and K.T. Leath. 1972. Genetic variance for reaction to five foliar pathogens in alfalfa. Crop Sci. 12:813-816.
13. Hill, R.R., Jr., K.T. Leath, and K.E. Zeiders. 1972. Combining ability among four-clone alfalfa synthetics. Crop Sci. 12:627-630.
14. Hill, R.R., Jr., M.W. Pederson, L.J. Elling, R.W. Cleveland, J.H. Graham, F.I. Frosheiser, and J.L. Starling. Comparison of expected genetic advance with selection on the basis of clone and polycross progeny-test performance in alfalfa. Crop Sci. 11:88-91.
15. Hill, R.R., Jr., and J.L. Rosenberger. 1985. Methods for combining data from germplasm evaluation trials. Crop Sci. 25:467-470.
16. Li, C.C. 1957. The genetic variance of autotetraploids with two alleles. Genetics 42:583-592.
17. Rowe, D.E. 1989. Competition thinning of alfalfa planted at three densities. Crop Sci. 29:1357-1361.
18. Rowe, D.E. 1988. Alfalfa persistence and yield in high density stands. Crop Sci. 28:491-494.
19. Rowe, D.E. 1987. Theoretical value of estimates of general combining ability in the autotetraploid crop. Theor. Appl. Genet. 73:537-541.
20. Rowe, D.E. 1985. Effects of random and non-random errors on phenotypic selection in autotetraploids. Theor. Appl. Genet. 69:317-323.
21. Rowe, D.E. 1982. Effect of gametic disequilibrium on selection in an autotetraploid population. Theor. Appl. Genet. 64:69-74.
22. Rowe, D.E., and R.R. Hill, Jr. 1984. Theoretical improvement of autotetraploid crops: Interpopulation and intrapopulation selection. U.S. Dept. of Agr. Tech. Bul. No. 1689, 32p.
23. Rowe, D.E., and R.R. Hill, Jr. 1984. Effect of gametic disequilibrium on means and on genetic variances of autotetraploid synthetic varieties. Theor. Appl. Genet. 68:69-74.
24. Rowe, D.E., and R.R. Hill, Jr. 1981. Inter-population improvement procedures for alfalfa. Crop Sci. 21:392-397.

k is selection pressure in terms of phenotypic variance; p and u are frequencies of desirable allele in improved population and topcross pollen parent, respectively; q is frequency of undesirable allele; r and n are number of replicates and number of plants in a replicate, respectively; a and ß are additve and digenic population effects, respectively; C is a component of the covariance due to selfing, details are found in reference (22); subscript W is variance in measurement of cloned materials treated the same, G is total genetic variance, e is error variance; a² with subscript designating a progeny is the variance among those progenies. This Table is copied from Table 11 of reference (22).

Previous Page