On the Evolution of Altruism by Kin Selection

Author(s): Carlo Matessi and Samuel Karlin

Source: Proceedings of the National Academy of Sciences of the United States of
America , Mar. 15, 1984, Vol. 81, No. 6, [Part 1: Biological Sciences] (Mar. 15, 1984), pp.
1754-1758
Published by: National Academy of Sciences

Stable URL: https://www.jstor.org/stable/23339

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

National Academy of Sciences is collaborating with JSTOR to digitize, preserve and extend
access to Proceedings of the National Academy of Sciences of the United States of America

This content downloaded from

181.222.111.44 on Mon, 15 Aug 2022 19:44:49 UTC
All use subject to https://about.jstor.org/terms
Proc. Natl. Acad. Sci. USA
Vol. 81, pp. 1754-1758, March 1984
Evolution

selection
On the evolution of altruism by kin
(cost of altruism/benefit of altruism/Hamilton rule)

CARLO MATESSI* AND SAMUEL KARLINtt

*Istituto di Genetica Biochimica ed Evoluzionistica, Consiglio Nazionale delle
arche,Ric
Via Sant'Epifanio 14, 27100 Pavia, Italy; and tDepartment of
Mathematics, Stanford University, Stanford, CA 94305

Contributed by Samuel Karlin, December 8, 1983

ABSTRACT A general model for the evolution of altruism pret the qualitative validity of the Hamilton rule to reflect
is formulated. Central to the model is a pair of local fitness the robust property that if an altruist gene is protected for a
population regime providing altruistic benefits to individual
functions, which prescribe the fitness of the altruist and selfish
phenotypes as functions of the composition of local groups into in a group having an average kinship relationship r, then th
which prereproductives are subdivided. When the local groups altruist gene would be protected for a corresponding popula-
are sibships or other kin groups, the model is one for kin selec-tion regime with closer group kinship relationship r' > r.
tion. Functions for cost and benefit of altruism are derived The concepts of quantitative and qualitative validity with
from the fitness functions. Conditions for evolution of altruism reference to the fixation event are defined analogously, mu-
are then determined in terms of cost and benefit. It is shown tatis mutandis.
that the Hamilton rule has quantitative validity only in the spe-In this paper we describe a flexible "fitness function" for-
cial case of linear fitness functions. Sufficient conditions are
mulation relevant to the population genetics of altruism (cf.
refs. 5, 6, and 12). The components of the model involve: (i)
found for qualitative validity of the Hamilton rule. Qualitative
violation of the rule is also possible. the delineation of the basic group structure specifying indi-
vidual relationships; (ii) the description of local fitness func-
The evolutionary basis of hierarchial status, cooperative for- tions depending on the group composition; (iii) the determi-
aging strategies, helping in brood rearing, signaling in prey- nation of population average fitness functions for the differ-
predator conflicts, and recognition and communication sys- ent phenotypes; and (iv) a pair of general benefit and cost
tems are discussed by some writers in terms of adaptive and functions, which depend on the group composition. The lat-
physiological correlates, where the role of kin selection ister functions are derived and do not enter as primitive quan-
considered central. Altruism might be involved in social in- tities.
teractions to the extent that they appear to increase the re-
productive success of some recipient individuals and de- Individual Fitness in Groups with Altruistic Interactions
crease the fitness of other conspecific members of the group,
the donors. Theoretical models of the population genetics of Consider an infinite population reproducing in discrete gen-
altruism and kin selection are many and varied (refs. 1-15; erations whose members belong to either of two phenotypic
see ref. 16 for a recent review and further references). classes: those who regularly perform some altruistic activity,
Hamilton (1) proposed the following rule as a basis for the the altruist A, and those who do not, the selfish individuals
evolution of a gene (or set of genes) promoting altruistic be- S. We assume that this polymorphism is controlled by an
havior: array of genotypes G1,G2,...,Gk underlying the phenotype
expression. The genotype-phenotype associations allow for
b/c > l/r, [1] partial penetrance such that P1,P2, . ,Pk denote the pene-
trance probabilities that the genotypes G1,...,Gk, respective-
where c is the cost to a donor ly, result in of
an altruist (A phenotype) and,behavior,
altruistic in the alternative b
benefit to a recipient per donor, event, a nonaltruist
and (S phenotype).
r is the appropriat
ficient of kinship among We assume that the
donors ethologyrecipients.
and of the species is such that
Rec
traditional values of r: for newborn individuals of
diploid generationr
sibs, t +=1 form
1/2;"appropriate"
for h
ploid sisters, r = 3/4; and groups
for of finite size N. The groups are
half-sibs, r considered
= 1/4. to be dis-
The evolutionary event of tinct from each other and persist
altruism canthrough
be the prereproduc- at
realized
eral levels. The usual criterion is that the altruistic allele be tive phase. Social interactions occur only among members of
protected (i.e., cannot go extinct when in low frequency).a group. When the prereproductive phase is complete, indi-
This will be referred to as the property of initial increase, I. viduals who successfully reach the reproductive stage dis-
A more compelling form of gene evolution entails the fixa- perse and mate at random.
tion event, F, signifying a total takeover by the allele in ques- We can describe a typical group composition at generation
tion once present in high frequency. t + 1 as follows: G1,G2,...,Gk are the possible genotypes,
We will speak of the quantitative validity of the Hamilton Z1,Z2,...,Zk are the corresponding numbers of the different
rule with respect to the property of initial increase provided genotypes, and Z1 + Z2 + ... +Zk = N is the group size. If
the following holds. There exists a measure of cost and bene- U(t) = [U1(t),U2(t),...,Uk(t)] is the vector of genotypic fre-
fit that may depend on the nature of gene expression andquencies of generation t, then generally Z = (Z1,Z2,...,Zk)
degree of penetrance but is independent of the kin-group is a random vector whose distribution depends on the mat-
structure such that the fulfillment of Eq. 1 signifies that aing pattern, group formation rule, and U(t). Let X =
newly introduced altruistic gene will increase in frequency. (X1,X2,...,Xk) be the vector whose coordinates are the num-
With respect to the property of initial increase, we inter- bers of A phenotypes of the various genotypes. As a first
approximation, we adopt the sampling rule that, given Zi, Xi
The publication costs of this article were defrayed in part by page chargehas a binomial distribution with parameters Zi,pi such that
payment. This article must therefore be hereby marked "advertisement"
in accordance with 18 U.S.C. ?1734 solely to indicate this fact. *To whom reprint requests should be addressed.

175L

This content downloaded from

181.222.111.44 on Mon, 15 Aug 2022 19:44:49 UTC
All use subject to https://about.jstor.org/terms
 Evolution: Matessi and Karlin Proc. Natl. Acad. Sci. USA 81(1984) 1755

each newborn of genotype Gi becomes an altruist with pene- pected fraction of altruists that survive (i.e., successfully
trance probability pi and a nonaltruist with probability 1 -reach
pi the reproductive stage) is
independent of the phenotypic expression of the other mem-
bers of the group. The aggregate number of altruists in a typi- N

cal group is X = Xl + X2 + ... + Xk. E (N)x(1 - o)N-"xfA(x)

Let fA(X) [fs(x)] be the probability that an A phenotype (S
phenotype) in a group that contains x altruists survives OA(0)to = E[fitness of Al ] =
NO
reproduce. The nature of the functions fs and fA relates to
the quality and intensity of altruism intertwined with ecologi-In a similar manner, we determine
cal and genetic influences.
Given U(t) = [U1(t), ..., U(t)], the vector of genotypic fre-
ss(O) = E[fitness of SIO]
quencies of the preceding generation, and by stipulating the N
mating structure and subsequent sampling scheme for gener-
ating groups of generation t + 1, the distribution of {Z,X} =x=0(xN)0(1 - O)N-X(N - x)fs(x)/(1 - O)N.
is determined. Let E(-IU(t)} denote the operation of taking
expectation of a random variable under this distribution. TheWe functions of AA(0) and bs(0) will be referred to as the
have Ui(t + 1) = E{XifA(X) + (Zi - Xi)fs(X)lU(t)}/N, expected fitness functions for an A and S phenotype, respec-
yielding a system of recurrence equations that'describes thetively, corresponding to a group that has an expected frac-
evolution of altruism in the population. tion 0 of altruists.
We shall explicitly consider the following kin-group struc-
tures. Derivation of Cost and Benefit of Altruism as Functions of
(i) Full Sibships. Members of each group are offspring ofGroup
a Composition
single mating couple. (In the haplodiploid case they are full
sisters.) The mating pattern and U(t) determine the frequen- Since Hamilton's rule (Eq. 1) is formulated in terms of cost
cy of each mating type. We shall concentrate on randomand benefit of altruism, most theoretical work on kin selec-
mating, but other patterns could be considered as well (e.g., tion is founded on specific models involving cost and benefit
assortative mating, partial consanguinity, and regular in- as primitive notions (usually constant parameters). Our gen-
breeding schemes). For a given mating type, the distribution eral method of fitness assignments does not deal directly
of Z is multinomial, with probability parameters determined with cost and benefit, so that a direct comparison of its re-
by Mendelian Laws. sults with Hamilton's prognosis would seem impossible. On
(i) Half Sibships. The members of each group are half- the other hand, from an operational point of view, cost and
sibs-i.e., they have a common mother but distinct fathers. benefit can only be defined as changes of fitness.
This can be regarded as a limiting case of multiple insemina-Consider a group with x altruists and N - x selfish individ-
tions as discussed (5, 12). Under random mating, given the uals, and, among these, a particular A individual, and call it
maternal genotype and the paternal gametic pool, the distri- I. If I could change its phenotype into S, it would have fit-
bution of Z is multinomial. nessfs(x - 1), but it actually has fitnessfA(x); thus, the cost
Other methods of group formation may include: (i) mixed to I of being an altruist is cost(I) = fs(x - 1) - fA(x).
sib/random groups where, with each parental pair, eachDefinition 1: The cost of altruism per donor in a group
member of the group is either a random offspring of the pair consisting of x A phenotypes and (N - x) S phenotypes is
or a newborn chosen at random from the population at large; c(x) = fs(x - 1) - f(x), x = 1,...,N.
(ii) groups determined by degrees of assortment based on If I, originally an A individual, could change its phenotype
genotype and/or phenotype; and (iii) groups based on popu- into S, each of the remaining (x - 1) A individuals in the
lation structure criteria. group would carry a fitness value fA(x - 1), and each of the
For a variety of group formation rules, the distribution of(N - x) S individuals would have a fitness fs(x - 1). Thus,
Z = (Z1,...,Zk), the various genotype numbers in a typical since I is an A individual, the marginal fitness contributed to
group, under proper conditioning, is multinomial with proba- each of the other As by each altruist is fA(x) - fA(x - 1),
bility parameters that either are constants or depend on thewhile the marginal fitness accruing to each S individual in
parental genotypic frequencies. More formally, we have the group is fs(x) - fs(x - 1). From this perspective, we
arrive at the following definition.
Definition 2: The cumulative benefit from altruism per do-
N! nor in a group with x A phenotypes and (N - x) S pheno-
Pr{Zl = Zi,...,Zk = Zk} = E P(X) ! ! ...kk,
x zl!...zk! types is b(x) = (x- l)[fA(x) - fA(X - 1)] + (N - x)[fs(x) -
fs(x - 1)], x= 1,...,N.
Definitions I and 2 define the marginal cost and benefit of
where P(X) is a probability distribution foraltruism A = (X1,... ,Xk) and
and recognize that these may vary with the size of
the components of X = (X1,...,Xk) indicatethe the expected
group and numbers geno-
of altruists in a group. It is natural to
typic frequencies in a given group. require, for a proper altruistic trait, that c(x) > 0, b(x) > 0 for
PROPOSITION 1. Under conditionally multinomial all x = 1,...,N. sam-
pling, the number of A phenotypes in a group, X = X1 + ...
+ Xk, follows conditional on X a binomialExpected distribution of N
Cost and Benefit Functions
trials with probability 0 = plhi + P2X2 +... + Pkk.
Obviously, 0 is the expected fraction of altruists
The cost and in a group
benefit functions described in Definitions I and
with sampling-segregation vector parameter 2 referX.to The popula-
the local group. For groups with expected propor-
tion at large can be viewed as stratified with
tion ofrespect
A phenotypesto equal
the to 0, we have the expected cost
parameter 0. Over the subpopulation of groups q3o
to a typical with
altruist ex- to such groups as y(8) = E[to-
belonging
pected fraction of altruists 0, a sampled group
tal cost ofis composed
altruism for all donorsl O]/E[number of donorsl 6] =
of X altruists and N - X nonaltruists following
xN=0o(N)x(1the distribu- Similarly, we determine 3(9),
- 0)N-xxc(x)/N0.
tion law Pr{X = xl0} = (mN) OX(1 - )N-x. the expected total benefit per altruist provided to the group,
Averaging with respect to all groups ofwhere
'So characterized
the group contains an expected proportion 0 of altru-
by an expected number NO of altruists, we find
ists, that the
as N=o0(C)0(l1 ex-
- 0)N-XXb(x)/N0.

This content downloaded from

181.222.111.44 on Mon, 15 Aug 2022 19:44:49 UTC
All use subject to https://about.jstor.org/terms
1756 Evolution: Matessi and Karlin Proc. Natl. Acad. Sci. USA 81 (1984)

We refer to y(0) and 3(0) as the expected cost and Conditions

benefit for Initial Increase and Fixation of a Gene
for Altruism
functions to an altruist and to recipients, respectively, of a
group with expected fraction 0 of altruists. For a proper al-
truistic trait, we have y(O),,3(0) > 0 for all 0 < 0 < It
1. is convenient to introduce the generalized average cost
It is straightforward to derive the identities and benefit functions for the interval (e1,:2), 0 ' < < 12 1,
namely,
y(0) = S(0) - A(0), P(0) = 0A(0) + (1 - 0)4s(0)
and QC(1,62) = j y(O)dO, B( ),12)= 2 _' 3(0)dO.

s(0) = s(0) + oy(0) + J 3(u)du Table 1 records the exact conditions ensuring initial increase
I and fixation F when the (A,S) polymorphism is controlled
by a single autosomal locus with two alleles, Ao and A1. The
- yo (u)du, A(0) = S(0) - y(0). altruist allele A1 is either fully dominant (D) or recessive (R)
It follows that the expected fitness functions {S(0),4)A(0)} and has penetrance p (0 < p - 1), so that the probabilities of
and the expected cost and benefit functions {y(0),,3(0)} expressing altruism for the three genotypes AoAo, AoA1, and
equivalently delimit the selective structure of these models. A1A1 are, respectively, 0, p, and p in the dominant case and
It is essential to emphasize that we cannot extract from the
0, 0, and p in the recessive case. The conditions of Table 1
expected cost and benefit functions in general a natural (in- can be derived from'those published in ref. 12, which are in
trinsic) cost-benefit set of marginal constant parameters. terms of a pair of functions A(w) and S(w) prescribing the
There are no such things. expected fitness of potential altruist and selfish genotypes,
A class of examples considers the one-function models of respectively, in a group where the expected proportion of
the form potential altruist genotypes is w. Because the functions 4OA
and 4s give the fitness of the altruist and selfish phenotypes,
s() = 1 + bh(0), A(0) = 1 + (1 - a)bh(O) - c [2] the relation between the two pairs of fitness functions, for
any fixed degree p of penetrance, is A(w) = p'A(pw) + (1 -
where a is a parameter O - a 1 and O ? c ? 1. For conve- p) s(pw), S(w) = os(pw) because, when the expected pro-
nience we factored out the parameter b > O to fix the normal- portion of genotypes which can express altruism is w, the
ization h'(0) = 1. In order for Eq. 2 to generate a meaningful expected proportion of altruistic phenotypes is 0 = pw.
set of benefit-cost functions, we'require h(O) to be mono- Inspection of Table 1 reveals that the conditions of initial
tone, increasing (h'(0),- 0). Obviously, h(O) apart from increase and fixation all have the form displayed in Eq. 3
translation and/or scale constants represents the fitness below (r is the relevant coefficient of kinship).
functions 'A(O) and 4s(0). RESULT 1. Defining f as below, the conditions for evolu-
The altruism fitness model of Eq. 2 can be given the fol- tion of altruism become
lowing interpretation: (i) all S individuals receive help from
A individuals; (ii) each A individual suffers an additive con- _ 31(i,e,p) 1 F 1
stant fitness loss, c; (iii) a fraction a of A individuals are not I(i,e,p) = I( ) r
(i,e,p) > - r and 1F(i
recipients of altruistic benefits; (iv) all recipients of altruism
gain an additive fitness increment bh(0), where 0 is the ex-
pected fraction of altruists in the relevant group. where i = 0, 1, or 2 index
The above model (Eq. 2) generalizes models I and II of model (O = haplodiploid s
Charlesworth (10), where h(O) = 0/(1 - 0), a = 1, and h(O) = half-sibs), the two-value
0, a = c, respectively. The choice h(O) - 0, a = 0 produces cessive or dominant gen
the Cavalli-Sforza-Feldman additive model (9). trance probability of an
The derived benefit and cost functions for the model of that evolution of altruism
Eq. 2 are, respectively, y(O) = c + abh(9) and 3(0) = b(land - fixation event, respe
a0)h'(O). sults in this paper and fur

ingene
Table 1. Conditions for initial increase and fixation of an altruist terms of cost and benefit f

Kind of Conditions for initial increase of altruist ger e Conditions for fixation of altruist gene
brood r (Instability of fixation of Ao) (Stability of fixation of Al)

O haploid R 3/4 4 3C(O,p/2) + y(p/2) 2B(0,p) + B(p/2,p) 4 2C(0,p) + C(p/2,p) + y(p/2)

B(O,p/2) > -- 3 4 3 3 4

1 diploid R 1/2 (/ C(O,p/4) + y(p/4) p2,p)>C

B(O,p/4) > 2 B(p/2,p)> 2
2 half-sib R 1/4
P(0) > 4y(0) B(p/2,p) > 4
4

0 haploid D 3/4 B(O,p/2) + 2 ) 4 ,+ 2B(0,p) ,p+ y(p/2) 4 3C(p/2,p) + y(p/2)

3 3 4
B(p/2,p)3> -4
'
1 diploid D 1/2 2) > 2 C(O,p/2) + y(p/2) C(3p/4,p) 4 +
B(3p/4,p) > 2

2 half-sib D 1/4 B(,p/2) > 4 2y(0) + C(O,p/2) + y(p/2)

.O,p) ~4 P3(p) > 4y(p)

uism; modelof
Model 0 refers to the haplodiploid model 1 refers to diploid sib altruism; model directed
sister-to-sister 2 refers to half-sibalt]
D (dominant).
altruism. The altruist gene expression is indicated by R (recessive) and

This content downloaded from

181.222.111.44 on Mon, 15 Aug 2022 19:44:49 UTC
All use subject to https://about.jstor.org/terms
 Evolution: Matessi and Karlin Proc. NatL Acad. Sci. USA 81 (1984) 1757

will appear in ref. 17; see also ref. 18.) dition fi'(1,R,p) > 2, which assures the property of initial
In each case : is a weighted average of the benefit func-increase of altruism in the diploid sib group, is also satisfied.
Then, because '(0,R,p) = '>(1,R,p), certainly '(0,R,p) >
tion /3(0). Similarly I is a generalized average of the cost
function y(O). To illustrate, in the haplodiploid model 4/3 obtains and altruism will satisfy the I condition in the
corresponding haplodiploid model. At the same time, it
1 p/2 could well be that X(1,R,p) = Xf'(2,R,p) < 4 so that, with
9'(O,R,p)= B(O,p/2) = - 13(0)d0,
this cost-benefit regime, altruism cannot evolve if it is di-
rected to half-sibs.
(0Q,R,p) = 3C(0,p/2) + y(p/2) =P/2
The following result underscores the very special nature of
,R,p) - = y(O)dm(O), the quantitative validity of the Hamilton rule.
RESULT 3. The quantitative validity of the Hamilton rule
under
where dm(O) is the probability measure the formulationcomposed
of Definition 4 (e.g.,of
Eq. 5)two
implies
parts, a density of total massthat3/4the fitness
spreadfunctionsuniformly
PA(O) and )s(O) are parallel
over linear
functions (equivalently, the
[0,p/2] plus a mass probability of 1/4 placed at the point p/2.benefit and cost functions are
constants)
The conditions of Eq. 3 in Result I and conversely.
resemble the Hamilton
criterion b/c > 1/r provided the A sufficient
undefined condition for the qualitative validity
constants b and of thec
Hamilton rule with
are replaced by appropriate probability respect to the initial
averages of the increasebene-
recessive
gene (IR) model consists
fit and cost functions. The temptation of the inequalities (1/4)X1'(2,R,p)
to interpret Eq. 3 as
simply a generalized form of (1/2)g('(1,R,p)
the Hamilton [half-sibs (IR) formula
r diploid (IR), designated
is mis- IR2
> IR1] and (1/2)Xf(1;R,p)
leading because the averaged benefit expression c (3/4)f'(0,R,p)
1 and [diploid (IR) >
cost
haplodiploidIndeed,
expression I' are model dependent. (IR), designatedthese
IR1 = IRO].quantities
We describe in Result 4 several sufficient conditions ex-
vary depending on the kin-group structure (i.e., the form of
pressed
ploidy), on the recessive or dominant in terms character
of the cost and benefitoffunctions
thethat ensure
gene
all of theof
underlying altruism, on the level above requirements for a recessive
penetrance, and gene on model.
the
definition of evolution of altruismRESULT 4.whether
(i) The qualitative measured
validity of the Hamilton
in rule
applies for initial
terms of initial increase or fixation increase and recessive gene expression
realizations.
(IR2 > IR1 > increase
RESULT 2. The condition for initial IRO) if or fixation for
p , 0 reduces to the common condition rp(O) > y(O).
Obviously the foregoing result is tantamount to the valid-
O) in [0,1/2] (the ' symbol
ity of Hamilton's rule under conditions of very stands for "increases
weak pene-
trance. '(y) not necessarily strictly"), [6a]
Definition 3: The qualitative validity of the Hamilton rule
is said to hold for evolution of altruism assessed by initial
increase I for a recessive gene expression R and penetrancey) in [0,1/2], and y(O) increases over (0,00
probability p if the following holds: and decreases thereafter [6b]

I'(1,R,p) > 2 implies '(0,R,p) > 4/3 and

[4] [the condition y(0)/02 , allows y(O) to in
JV'(2,R,p) > 4 implies X'(1,R,p) > 2. strongly].
(ii) The qualitative validity of the Hamilton rule with re-
Analogous to Eq. 4, there exist versions of the qualitative spect to the fixation event and a recessive gene (FR2 a FR1
validity for I and dominant gene expression and the corre- a FRO) applies if
sponding relations underlying the fixation F criterion for
evolution of altruism.
With respect to quantitative validity of the Hamilton rule,
/(e0)/y(O) I in [0,1] or y(O) T in [0,1/2]. [7]
we state the following definition.
Definition 4: For a given criterion of evolution of altruism (iii) The simultaneous qualitative validity of the Hamilton
(say I or F) and a given gene expression (D or R), we say rule with respect to both I and Ffor a recessive gene ensues
that the quantitative validity of the Hamilton rule holds pro- if
vided the general benefit to cost ratio functions IX(p) of Eq. 3
are independent of the kin groups. p3(0)/y(0) T , y(O) t, but y(O)/02 4 on [0,1/2]. [8]
There are several natural explicit versions reflecting quan-
titative validity. The quantitative validity of the Hamilton There are corresponding results for the case of dominant'
rule for the criterion of initial increase I and a recessive gene gene expression.
expression R requires the equations We conclude this section by considering the qualitative
validity of the Hamilton rule for the generalized fitness func-
f'(O,R,p) = Jf'(1,R,p) = 2C(2,R,p) for all p. [5] tion model of Eq. 2.
RESULT 5. (i) For the class offitness models ofEq. 2 inde-
In the presence of Eq. 5, we can unambiguously refer to pendent of the determination ofh(x) as long as h'(x) - 0, the
any of the terms Jt'( ,R,p) as the benefit-cost ratio function. qualitative validity of the Hamilton rule appliesfor FR (fixa-
This is not a ratio of constants but depends on p and the tion realizations and recessive gene expression) and for the
specifications I and R. ID cases.
There are corresponding versions of quantitative validity (ii) For the initial increase recessive gene case (IR) it is
for the evolutionary criterion I and dominant gene expres- sufficient that h(x) be convex increasing in order that the
sion and for F coupled to either recessive or dominant gene qualitative validity of the Hamilton rule holds.
expression.
In particular, for h(0) = and a = 1, the actual
It is clear, under the conditions of the quantitative validity 1-0
of the Hamilton rule, that altruism will evolve more easily Charlesworth model I (10), the qualitative va
for a closer kin-group structure. Indeed, suppose the identi- prevails for all criteria of evolution of altruism
ties of Eq. 5 hold for the given gene expression, and the con- of gene expression.

This content downloaded from

181.222.111.44 on Mon, 15 Aug 2022 19:44:49 UTC
All use subject to https://about.jstor.org/terms
1758 Evolution: Matessi and Karlin Proc. Natl. Acad. Sci. USA 81 (1984)

Violation of Qualitative Validity of Hamilton's Rule ear fitness functions of the proportion of altruists in the rele-
vant group.
x
(iii) In assessing the circumstances for the qualitative va-
The specification h(x) = in Eq. 2 [this h(x) is concave
1+ mx lidity of the Hamilton rule a qualitative dictum emerges re-
increasing rather than convex increasing] with volving on the following contrasting themes:
m large
enough falsifies the qualitative validity for the model IR at
complete penetrance or near complete penetrance. The same "few altruists are better than many" [I]
example produces violation of the Hamilton rule in the FD
(fixation dominant) context. versus

Discussion
"many altruists are better than few." [II]

The Hamilton rule (Eq. 1) and the principle of "Maximiza- The qualitative validity of the Hamilton rule for the criterion
tion of Inclusive Fitness" have been proposed as an intrinsic of initial increase tends to apply in the circumstance of
construct for the evolution of altruism by kin selection. It istheme II. This, in essence, occurs in the case of decreasing
contended that the closer coefficient of kinship for haplodi- cost function (i.e., cost per altruist diminishes), whereas a
ploid sisters compared to diploid siblings allows more oppor- larger fitness benefit accrues to the recipients as the fraction
tunities for the fulfillment of the condition b/c > l/r and, of altruists in the group increases. The same conclusion ob-
concordantly, genes for altruism are more easily established. tains when ,3(0)/y(0), the "efficiency" of doing altruism, in-
The prototype case is the social hymenoptera insects with creases with 0.
sterile castes.
In the purview of the criterion of fixation, the qualitative
The simplicity of the Hamilton rule is appealing. Howev- validity of the Hamilton rule conforms roughly to the dictum
er, it is undeniable that behavioral traits by their very natureof theme I such that few altruists, rather than many, contrib-
are intrinsically complex. They undoubtedly involve multi- ute more effectively to the fitness endowments of both A
factorial inheritance coupled to a myriad of gene-environ-and S individuals. This applies to the case where y(0) in-
ment interactions and learning and cultural transmission fac- creases and 3(0) decreases or 3(0)/y(O) decreases.
tors. Even in the context of one-locus models, the prediction
of Eq. I is defiantly simple. Notice that the quantities b, c, This work was supported in part by National Institutes of Health
Grant 5RO1 GM10452-20 and National Science Foundation Grant
and r appear to be independent of the degree or form of gene
MCS79-24310.
expression (e.g., level of dominance and penetrance), geno-
type-phenotype determinations, mating pattern, the nature1. Hamilton, W. D. (1964) J. Theor. Biol. 7, 1-16; 17-52.
and distribution of recipients and donors within and between2. Maynard Smith, J. (1965) Am. Nat. 99, 59-63.
"groups," group size, sex ratio, and other population-struc- 3. Eshel, I. (1972) Theor. Popul. Biol. 3, 258-277.
ture covariates. 4. Boorman, S. A. & Levitt, P. R. (1973) Proc. Natl. Acad. Sci.
We highlight the salient conclusions and some interpreta- USA 70, 187-189.
tions of our results: 5. Levitt, P. R. (1975) Proc. Natl. Acad. Sci. USA 72, 4531-4535.
(i) For a given kin-group structure, the conditions imply- 6. Matessi, C. & Jayakar, S. D. (1976) Theor. Popul. Biol. 9, 360-
387.
ing I and F all have the form of Eq. 3 (Result 1), where X is a
7. Cohen, D. & Eshel, I. (1976) Theor. Popul. Biol. 10, 276-302.
generalized benefit-to-cost ratio, the numerator being an ap-
8. Wilson, D. S. (1977). Am. Nat. 111, 157-185.
propriate average of the benefit function and the denomina-
9. Cavalli-Sforza, L. L. & Feldman, M. W. (1978) Theor. Popul.
tor an appropriate average of the cost function. The explicit Biol. 14, 268-280.
averages depend on the criterion for assessing evolution of 10. Charlesworth, B. (1978) J. Theor. Biol. 72, 297-319.
altruism, on the nature of gene expression, and on the kin- 11. Wade, M. (1979) Am. Nat. 113, 399-417.
group structure. If further factors are incorporated into the 12. Boorman, S. A. & Levitt, P. R. (1980) The Genetics of Altru-
model (e.g., more population- and group-structure parame- ism (Academic, New York).
ters, preferential mating pattern, multiple phenotypic grada- 13. Uyenoyama, M. K. & Feldman, M. W. (1981) Theor. Popul.
tions, genetic modifiers) we would expect XR to reflect these Biol. 19, 87-123.
factors. 14. Uyenoyama, M. K., Feldman, M. W. & Mueller, L. D. (1981)
Proc. Natl. Acad. Sci. USA 78, 5036-5040.
(ii) The concept of quantitative validity of the Hamilton 15. Engels, W. R. (1983) Proc. Natl. Acad. Sci. USA 80, 515-518.
rule requires the existence of a generalized benefit-to-cost 16. Michod, R. E. (1982) Annu. Rev. Ecol. Syst. 13, 23-55.
ratio function that is independent of the kin-group structure. 17. Karlin, S. & Matessi, C. (1984) Theor. Popul. Biol., in press.
Result 3 shows that this is only realized in the special case of 18. Karlin, S. & Matessi, C. (1983) Proc. R. Soc. London Ser. A
constant benefit and cost functions equivalent to parallel lin- 13, 327-353.

This content downloaded from

181.222.111.44 on Mon, 15 Aug 2022 19:44:49 UTC
All use subject to https://about.jstor.org/terms