Proc. Nat. Acad. Sci. USA Vol. 72, No. 7, pp. 2761-2764, July 1975 Genetics Distribution of allelic frequencies in a finite population under stepwise production of neutral alleles* (population genetics/frequency distribution/electrophoretically detectable alleles) MOToo KIMURA AND TOMOKO OHTA National Institute of Genetics, Mishima, 411 Japan Contributed by Motoo Kimura, April 25, 1975 ABSTRACT A formula for the distribution of allelic In particular, Ho represents the average homozygosity or the frequencies in a finite population is derived assuming step- expectation of the sum of squares of allelic frequencies. wise production of multipfe alleles. Monte Carlo experiments Since the effective number of alleles (ne) is given by the re- were performed to check the validity of the formula, and ex- cellent agreement was obtained between theoretical distribu- ciprocal of Ho, formula 1 follows immediately from 3. Fur- tion and experimental results. The formula should be useful thermore, the expectation of the product of frequencies of for analyzing genetic variability in natural populations that adjacent alleles is can be detected by electrophoretic methods. 1 + 4NeV -r11 + 8Nev For the purpose of analyzing genetic variability that can be C = [51 detected by electrophoretic methods, we recently proposed 4NeV F1 + 8Nev a model of stepwise production of alleles (1, 2). In this model, it is assumed that the entire sequence of allelic states Note that these results are concerned with the second mo- can be expressed by integers (..., A,1, AO, A1, .. .), and ments of the distribution of allelic frequencies rather than that if an allele changes states by mutation it moves either the distribution itself. one step in the positive direction or one step in the negative However, in order to assess, by detailed analyses of obser- direction in the allele space (Fig. 1). As compared with the vations, the role of mutation and random drift for the main- conventional model of Kimura and Crow (3) which assumes tenance of protein polymorphisms, knowledge on the actual that every mutation leads to a new, not preexisting allele, form of the distribution is required. Although considerable this model has a feature that mutations are to some extent information can be gained on the nature of the distribution recurrent; less frequent alleles in a population tend to be through careful Monte Carlo experiments (4), it is much produced repeatedly from mutation of more frequent adja- more desirable to derive the distribution analytically. The cent alleles. Analytical treatment of this model is much more purpose of the present paper is to show that under a simpli- difficult than that of Kimura and Crow, but we have ob- fying assumption (as expressed by Eq. 10 below), a distribu- tained, using diffusion equation methods, the formula for tion can be obtained that satisfies known relations 3 and 5, the effective number of selectively neutral alleles main- and also gives excellent fit to Monte Carlo experiments for tained in a finite population (2). Namely, if v is the mutation 4Nev up to unity. rate per locus per generation such that the mutational changes toward the positive and the negative directions BASIC THEORY occur with equal frequencies as shown in Fig. 1, and if Ne is Let us consider a random mating population of N diploid in- the effective size of the population, then, assuming selective dividuals, and let Ne be the effective size of the population neutrality of mutation, the effective number of alleles at which may be different from the actual size N. We assume equilibrium is given by that under mutation and random drift a statistical equilibri- um is reached with respect to distribution of allelic frequen- = 1 + [1] cies. We shall designate this distribution by CF(x) such that ne 8NeV* 4.(x)dx represents the expected number of alleles whose More generally, we have shown that if Ck is the expectation frequencies in the population are in the range (x - x + dx). of the product of the frequencies between two alleles that Consider an allele (say Ao) whose frequency in the popu- are k steps apart in the allele space, then lation is x. Let x-i and xi be the frequencies of alleles (A-1 and A1) that are adjacent to Ao. Then, under stepwise pro- Ck = H0X I, [2] duction of alleles as shown in Fig. 1, the mean and the vari- ance of the change of x per generation are given respective- where ly by Ho = 1/1+ 8Nev [3] V and M = -ux + - 2 E(x-,. + xljx) [6] X, = (1 + 4NeV - 1 + 8Nev)/(4Nev). [4] and Vsx = x(l - x)1(2N, ), [7] * Contribution no. 1057 from the National Institute of Genetics, where v is the mutation rate and E(x_1 + xilx) is the condi- Mishima, Shizuoka-ken 411 Japan. tional expectation of the sum of the frequencies of the two 2761 Downloaded by guest on October 4, 2021 2762 Genetics: Kimura and Ohta Proc. Nat. Acad. Sci. USA 72 (1975) V/2 V/2 V/2 V/2 Thus, the required distribution may be expressed in the fol- lowing form: . A-2 A1 A0 Al A2 .. + 0 1 = F( + 1) (1 - - - I [15] FIG. 1. Diagram showing the model of stepwise productic)n of 4(x) + x)'t Xl alleles. F(a)F(h= 1) where a = 4Nev, 03 = 2Ne,,vb and adjacent alleles given that the frequency of the allele in question is x. + 4Nev - + b =l1 V1 8Nev If E(x_1 + xllx) can be expressed as a known function of [16] x, then CF(x) may be obtained by applying Wright's (5) for- 2Nv(l 1 + 8NeV - 1) mula for the steady-state gene frequency distribution, that We shall now show that with this distribution and the is, by above assignment of constant b in Eq. 10 we can derive the correct value for the expectation of the product of frequen- F(x) = V5xexp{2 f~5dx [8] cies between adjacent alleles. In terms of the present distri- bution, this is given by where C is a constant determined such that the sum of a llel- ic frequencies in the population is unity: Cl = 2 xE(x-, + xJxX4(x)dx and we f xF(x)dx = 1. and, noting Eqs. 9, 10, 13, 16, get 0 [9] b C' Note that this condition is different from the one originally Cl = x(1 - x)4?(x)dx used by Wright to determine the constant. The reason for - this is that in his case the distribution represents probability = b(1 - Ho) = 1 1_+8Ne) density, while in our case 4(x) represents density of the ex- pected number of alleles (see also refs. 3 and 6). 1 + 4Nev - 1/ + 8Nev Let us now assume that E(x-l + x1lx) can be expressed, with sufficient accuracy, as a constant fraction of the total 4NeV C1 + 8Nev frequencies of the remaining alleles so that This agrees with Eq. 5, the result obtained earlier by us (2) + = b(1 - x), [10] using an entirely different method. Furthermore, by letting E(x-l xjx) f = xn(n > 2) in EjL(f)j = 0 in Ohta and Kimura's (ref. 2) where b is a constant that may depend on v and Ne but not formulation, we find that 4(x) must satisfy the relation: on x. Substituting Eq. 10 in Eq. 6 we get (n - 1) { xn - I(x)dx - fxn4?(x)dx} M = -vx + 2vb(1 - x), [6a] -4NeV {fxn4?(x)dx and, this together with Eq. 7, allows us to obtain the distri- bution by using formula 8. Thus, we obtain - 2fE(x_1 + xIIx)xn -'4(x)dx} = 0, (X) =C(1 - X)' x - 1 [1I] and we can show, in fact, this is satisfied by Eq. 15, assum- where a = 4Nev and f = 2Nevb. In this formula, constant C ing Eq. 10. We can also show that as NeV gets small, b ap- is determined by using condition 9, and we get proaches unity so that 4(x) approaches the distribution given by Eq. 11 but with a = 4Nev and :3 = 2Nev. On the F(a + ,B + 1) other hand, it is known (see ref. 6) that this form of distribu- F(a)F( + 1) tion represents the case in which the entire allelic space con- sists of three allelic states with stepwise production of muta- In order to determine another constant b, we note that the tions. This is reasonable since when NeV is small, the number expectation of the sum of squares of allelic frequencies is Ho of different alleles contained at any moment within a popu- = 1/x/T + NeV as given in Eq. 3. In terms of the present lation seldom exceeds three, so that the triallelic space distribution, we have should be sufficient for describing the distribution. Inciden- tally, we note that b can be expressed in the form: Ho = x24(x)dx = ( + 1)/(a + 3 + 1) 2 [16a] [13] ne- 1 and therefore, by equating this with 1/x/]7+ WNe, we ob- Since the effective number of alleles excluding the allele tain under consideration (i.e., Ao) is ne - 1, of which two are in the states adjacent to Ao, and, since X1 = C,/Ho represents, / = (a + 1 - V1+ 8NeV)/(l + 8Nev - 1). in a sense, correlation between adjacent alleles, we think that [14] b given by Eq. 16a allows us a very natural interpretation. Downloaded by guest on October 4, 2021 Genetics: Kimura and Ohta Proc. Nat. Acad. Sci. USA 72 (1975) 2763 Table i. Comparison of na between Kimura-Crow and stepwise production models N 5 x 102 103 5 x 103 104 Ne v K-C Step K-C Step K-C Step K-C Step 0.01 1.274 1.256 1.301 1.280 1.366 1.335 1.394 1.358 0.025 1.675 1.578 1.745 1.627 1.906 1.736 1.975 1.780 0.05 2.324 1.994 2.462 2.068 2.784 2.220 2.923 2.280 0.1 3.557 2.562 3.834 2.649 4.478 2.816 4.755 2.874 0.25 6.908 3.434 7.601 3.501 9.210 3.604 9.903 3.633 Comparison of the average number of alleles (n.) between the Kimura-Crow model (K-C) and the model of stepwise production of alleles (Step), under various combinations of N and NeV, where N and Ne are respectively the actual and the effective numbers of the population, and v is the mutation rate.