RECURRENT SELECTION AND OVERDOMINANCE 471 



Now if we are convinced that overdominance is not very important and 

 that, perhaps for other reasons too, selection for general combinability will 

 eventually win, or at least not lose, we may proceed at once with recurrent 

 selection for general combinability to render hybrid corn obsolete. Some of us 

 may find it necessary to include an inbreeding interphase between cycles. 

 Breeders of livestock may as well return to improvement of pure breeds by 

 progeny testing. We will run these pilot tests merely for the sake of verifica- 

 tion. 



But if it should seem likely that recurrent selection for specific combina- 

 bility may win, the breeder of livestock may begin now with recurrent recip- 

 rocal selection for specific combinability. For my part, I would choose two 

 crossbreds for the start and would begin mild inbreeding in one of them which 

 would become the stud herd. On one side of this is the Comstock plan with 

 no inbreeding in either herd. On the other side we might choose a line with 

 50 per cent inbreeding at the start and practice reciprocal selection along with 

 continued mild inbreeding. Evaluation of these alternatives of the reciprocal 

 plan with small laboratory animals, along with the other two main plans, 

 would be of considerable interest theoretically. The cost might be minute in 

 comparison with the total of wasted effort in current breeding practices. 



Recurrent selection for general combinability without the inbreeding in- 

 terphase is a fairly obvious technic which has been employed and described 

 variously. The first discussion of it from the overdominance viewpoint with 

 the restriction against selection for specific combinability was that of Hull 

 (1946b). Since then I have continued to urge parallel tests with fast breeding 

 species as pilot experiments. Recurrent selection for superior homozygotes is 

 proposed here for the first time, I believe. 



Reciprocal selection for specific combinability was a counter proposal to 

 me of several corn breeders in 1944 and later, when I proposed selection in a 

 crossbred for combinability with a fixed tester, a homozygous line or Fi of 

 two homozygous lines. 



For simplicity of illustration we may consider a 4-factor example with gene 

 frequency in a homozygote or gamete (v or w) taking values, 0, |, f , f , |. 

 Gene frequencies intermediate to these values may occur in heterozygotes 

 and in whole populations. Let us take ^ = 2 for the degree of dominance as 

 suggested roughly for corn yield by estimates reported here. Then regression 

 of ofi'spring phenotype on gene frequency of parent in any column of the 

 (5 X 5) Mendelian checkerboard is bp = ^(3) — 2v, where v is gene frequen- 

 cy of the common parent of the column. Substituting the five values of v pro- 

 vides the five values of bp, 1^, 1, |, 0, — |, for the five columns. Heritability 

 changes from positive to negative where v ^ (I -\- k)/2k = f . These values 

 of bp for the given values of v are the same for any number of loci. In any 

 case the zone of near-zero heritability for one locus is relatively broad on both 

 sides of the critical value of zero. Reciprocal selection between two crossbreds 



