470 FRED H. HULL 



Present-day corn breeding is done in three steps: selection among inbreds 

 based on their own phenotypes; selection among inbreds for general combina- 

 bility ; selection among specific Fi's of the remaining inbreds. These steps are 

 the three processes of the preceding paragraph. The corn breeder applies the 

 three processes in the order named to the same stock, then recombines the 

 elite lines and begins the cycle again. The present proposition is to apply the 

 three processes separately to parallel stocks, and thus attempt to learn which 

 ones are primarily responsible for superior hybrids. 



RECURRENT SELECTION AMONG HOMOZYGOTES 



This process can be done effectively enough in corn, perhaps with S2 lines. 

 Two selfings would amount statistically to reducing the degree of dominance 

 to one-fourth of the original value. One-half of the Si lines could be discarded 

 in the first comparison. About fifty S2 lines should be retained in the recom- 

 bination. Selection within ear-rows should be rigidly excluded. 



There is no reason to suppose that a physiological barrier would be reached 

 short of the level of elite hybrids. Recurrent selection towards an extreme has 

 been very effective with many characters where not much dominance is ap- 

 parent. In noted cases no limit of genetic variance has been reached. What 

 genetic limit might be reached with vigor or yield genes of corn when the con- 

 fusion of dominance is artificially eliminated is to be explored. Theoretically, 

 this process of recurrent selection should be much superior to any non-recur- 

 rent selection among gametes or doubled haploids. 



RECURRENT SELECTION FOR GENERAL COMBINABILITY 



Strictly, the tester should be the variety. So plants or S„ lines are to be 

 testcrossed with several plants of the variety. The So plant must be selfed at 

 the same time. Parents of elite testcrosses are recombined into an improved 

 variety which becomes the tester for the next cycle. If gene frequency of the 

 variety is improved to approach (1 + k)/2k, where ^ > 1, heritability will 

 approach zero and the variety mean its maximum. If pseudo-overdominance 

 from repulsion linkage is important the equilibrium may advance to higher 

 levels as recombinations occur. But, aside from that, we have now no experi- 

 mental verification of a selection equilibrium, and a test would seem desir- 

 able. Concurrent selection for specific combinability should be strictly avoid- 

 ed in this test. 



RECURRENT SELECTION FOR SPECIFIC COMBINABILITY 



This process has been adequately described both here and earlier (Hull, 

 1945a). From the theoretical viewpoint it would be best to use a homozygous 

 tester and avoid selection within the crossbred except that based on testcross 

 performance. The purpose is to determine first how much specific combina- 

 bility may be accumulated in early cycles and eventually to determine where 

 this process may reach physiological or genetic limits. 



