W!llHlWNUHHWNW!HWHHWWI

   

A THEORY OF GROUP SELECTION
DISSERTATION FOR THE DEGREE OF m
MICHIGAN STATE UNIVERSITY

DAVID SLOAN WILSON

1976

 

 

This is to certify that the

thesis entitled

A Theory of Group Selection

presented by

David Sloan Wilson

has been accepted towards fulﬁllment
of the requirements for

Ph.D. degreein Zoology

// Mel/w/

Majproessf

 

 

DateW/

0-7639

ABSTRACT

A THEORY OF GROUP SELECTION

by

David Sloan Wilson

In organisms possessing a dispersal phase the processes of
mating, competition, feeding and predation are often carried out in
spatial areas (trait—groups) smaller than the boundaries of the
deme. A simple model shows that this can lead to the selection of
'altruistic' traits that favor the fitness of the group over the
individual. The extent of group selection that occurs depends
mainly on the variation in the composition of genotypes between
trait-groups. The traditional concepts of group and individual
selection are seen as two extremes of a continuum, with systems

in nature operating over the interval in between.

A THEORY OF GROUP SELECTION

BY

David Sloan Wilson

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Zoology

1976

ACKNOWLEDGMENTS

I am indebted to D. J. Hall, Conrad Istock, W. C00per,
E. E. Werner, E. O. Wilson, G. C. Williams, H. Caswell, F. M.
Stewart, B. Levin, and the Ecology group at Michigan State for

helpful criticism and discussion.

ii

LIST OF FIGURES

INTRODUCTION

TABLE OF CONTENTS

A THEORY OF GROUP SELECTION

CONCLUSIONS

BIBLIOGRAPHY

iii

Page

iv

LIST OF FIGURES

Figure Page

1. The entire set of traits giving fitness changes to
the donor and to each recipient (fr) . . . . . . 4

2. Illustration of the group selection process . . . . 5

iv

INTRODUCTION

The theory of group selection became controversial in 1962
when V. C. Wynne-Edwards postulated that animals hold their density
below carrying capacity to avoid overexploiting their resources.
Unfortunately Wynne-Edwards' ideas on how such a 'management‘ policy ‘
could evolve were vague and unconvincing.

Since then one of the greatest impediments to the development
of the theory has ironically, been its enormous appeal. Against a
Darwinian landscape of competition and selfishness, it pr0poses
cooperation and altruism. People want to believe it, and evolutionary
biologists, instead of exploring it as an exciting possibility, have
until recently felt obligated to hold back a flood of uncritical
acceptance with ominous reminders about parsimony.

During the last few years, however, a flush of recent models
indicate that group selection is being approached in a more objective
spirit (e.g., Levins 1970, Boorman and Levitt 1973, Levin and Kilmer
1974, Gilpin 1975). Most are refinements of Wynne—Edwards' initial
conception of interdemic selection--a cluster of small groups,
completely isolated except for a trickle of dispersers. Within each
group natural selection promotes increased resource utilization, even
to the point of overexploitation. Groups that overexploit go extinct,

however, so given a variation in the composition of genotypes between

groups (created by genetic drift and founder effects) differential
extinctions can create a form of 'group' selection promoting resource
management. These models conclude that although group selection can
be a significant force in a mathematical sense, the conditions pre-
supposed are rarely met in nature.

Even though these models are weak, this does not mean that
group selection is a weak force in nature. Traditional evolutionary
theory still labors under an unreasonable assumption of spatial
homogeneity of genotypes. Interdemic variation is one way to relax
the assumption, but it is not the only way.

This thesis presents a simple model of group selection based

on small scale, intrademic spatial variation of genotypes. The

 

concept of intrademic spatial variation at first sounds contradictory,
for a deme is by definition a spatially homogenous mating population.
However, the concept is realistic due to a ubiquitous feature of
organisms: Almost all plants and animals concentrate their movements
in a brief dispersal stage, and it is this stage that sets the size
of the deme. It follows that all traits manifested during the non-
dispersal stages must occur within populations that are smaller than
the deme. In short, as far as the manifestation of traits are con-
cerned, the deme is not only a population of individuals, but also a
population of groups.

When this concept is applied to traditional models, it can
be shown that both individual and group selection represent absurd
extremes of a continuum, with populations in nature Operating over

the interval in between.

Reprinted from

Proc. Nat. Acad. Sci. USA
Vol. 72, No. 1, pp. 143—146, January 1975

A Theory of Group Selection

(altruism/natural selection/structured demes/spatial heterogeneity)

DAVID SLOAN WILSON“

Department of Zoology. Michigan State University. East Lansing, Mich. 48823

Carmnunicated by Edward 0. lVilson, October 2.3, 1974

ABSTRACT In organisms possessing a dispersal phase
the processes of mating, competition, feeding, and pre-
dation are often carried out within “trait-groups,” deﬁned
as populations enclosed in areas smaller than the bound-
aries of the deme. A simple model shows that this can lead
to the selection of “altruistic” traits that favor the ﬁt-
ness of the group over that of the individual. The extent
of group selection that occurs depends mainly on the varia-
tion in the composition of genotypes between trait-groups.
The traditional concepts of group and individual selection
are seen as two extremes of a continuum, with systems in
nature operating over the interval in between.

 

Most theories of group selection (1—6) postulate many groups
ﬁxed in space, with exchange by dispersers between groups.
Within groups individual selection operates; an “altruistic”
trait can thus only become ﬁxed by genetic drift. This re-
quires the groups to be small, and dispersal between groups
must be slight to prevent the reintroduction of “selﬁsh”
individuals. The “altruistic” groups could then serve as a.
stock for the recolonization of selﬁsh groups that go extinct.
See ref. 7 for a review.

The recent models of Levins (4), Boorman and Levitt (5),
and Levin and Kilmer (6) make it plausible that this process
can occur—the main question is to what extent the conditions
for its operation (small group size, high isolation, high ex-
tinction rates) are met in nature. The current consensus is
that the proper conditions are infrequent or at- least limited to
special circumstances such as the early stage of colonization
of many populations (7).

This paper presents a theory of group selection based on a
different concept of groups, perhaps more generally met in
nature.

Most organisms have a dispersal stage—the seeds and
pollen of plants, the post-teneral migratory phase of adult
insects (8), the larvae of benthic marine life, the adolescents
of many vertebrates. This means that individuals are spatially
restricted during most of their life cycle, with the exception of
their dispersal phase, when what was previously a boundary is
easily transcended. As an example, a caterpillar is restricted
to one or a few plants, but as a butterﬂy it spans whole fields.

Evolution’s most easily conceived population unit is the
deme, and it is determined by the movement occurring during
the dispersal phase. Yet most ecological interactions, in terms
of competition, mating, feeding and predation are carried out
during the nondispersal stages in the smaller subdivisions,

 

“ Present address: The Biological Lalmratorics, llarvard Uni-
versity, Cambridge, Mass. 02138. After March 197.3: l.)epartment
of Zoology, University of Washington, Seattle, Wash. 9319.3

143

which I term “trait-groups.” In some cases the trait-groups
are discrete and easily recognized, such as for vessel-inhabiting
mosquitoes and dung insects. In other cases they are con-
tinuous and each individual forms the center of its own trait-
group, interacting only with its immediate neighbors, which
comprise a small proportion of the deme. Two examples are
plants and territorial animals. The following model treats
only the discrete case, but the results can be generalized.

In order to determine if a heritable trait manifested within
the trait-group will be favored in selection, the effect of that
trait on relative ﬁtnesses within the (rail-group must be
modelled, and the relative ﬁtnesses for the deme obtained by
taking the weighted average over all the trait-groups in the
deme. Traditional models of selection neglect this; i.e., they
assume that the trait-group equals the deme.

THE MODEL

Consider a single, haploid, randomly intermixing trait-group
of organisms, composed of two types of individuals, A and B.
These differ by only a single heritable trait, such as feeding
rate, aggressiveness, or behavior under the threat of preda-
tion. Because the two types are identical in every other
respect, they will have the same “baseline” ﬁtness, and
diﬁercnces can be attributed solely to the effect of the trait.
Haploidy and baseline fitness are of course artificial for most
populations. They are used to simplify the argument, and the
fundamental conclusions are not dependent upon them.
Space does not permit a fuller treatment, which will be pre-
sented elsewhere.

By manifesting its trait, every A-individual changes its
own ﬁtness and often the fitness of the other animals in the

'trait-group by a certain increment or decrement. Call the
individual manifesting the trait the donor, and all those
affected by it (both A and B types) the recipients. These.
terms are commonly used for altruistic social behaviors, but
here they are applied to all traits. For instance, an animal
with a higher feeding rate deprives its neighbors of food that
otherwise would be available to them. A positive ﬁtness
change is thus bestowed upon the donor and a negative fitness
change to the recipients, even though the animals never
interact behaviorally with each other.

Graphically, any trait can be portrayed as in Fig. 1. Each
point on the graph represents a trait with its fitness change to
the donor (fa) and to each recipient.» (L). In the example just
given the trait would lie somewhere in the fourth quadrant
(fd positive, f, negative). As another example, a warning cry
might decrease the fitness of the caller (fa negative) and in-
crease the fitness of those that bear the cry (I, positive),

144 Zoology: Wilson

 

+
2 f, I
\“\“
§
\
‘
- \‘ ‘ fd
“‘\\ +
\
‘
\
“
m
3 4

 

FIG. 1. The entire set of traits giving ﬁtness changes to the
donor (fa) and to each recipient. (f,). Any point. to the right of the
diagonal solid line is selected for; in the position shown the solid
line represents the traditional concept of individual selection (f4 >
f,). Any point to the right of the broken line increases the ﬁtness
of the group (fa > —(N — 1)f,). As the variation in the com-
position of trait-groups increases, its effect on the selection of
traits is to rotate the solid line until it. is coincidental with the
broken line (arrow).

placing the trait somewhere in the second quadrant. In this
model it is assumed that f, is the same for both A and B types.
While this will often be false in nature, it serves as a foundation
for more realistic elaborations.

If N is the total trait-group size and a,b the proportions of
the A and B types, respectively, the average per capita ﬁtness
changes resulting from the trait can easily be calculated.

per capita ﬁtness change to the A-typc = L: + N(a — 1.,»"N)f,
per capita ﬁtness change to the B-type = Na],

Traditional selection models assume that. the trait-group
equals the demo. In this case the model is sufficient as stands
and the .-\-individuals are. selected for only if they have a
higher per capita fitness change than the B-imlividuals.

fa + .\'(a - l,.-".\')f, > No], [l]
fa > .Vaf, — .\'(a — 1,5.\')f,
fd > fr [2]

Expression [2] is the traditional concept. of individual selec-
tion, i.e., the trait. innst‘. give the individual possessing the
fruit :1 higher relative fitness than the individuals not possess-
ing the trait. This is portrayed in Fig. l by the solid line.
Any point to the right of this line will be selected for by
individual selection.

llowever, the -\-frait increases the fitness of the trait-
group only if:

f.) + (.\' — Hf, > o
1}, > —«-.\' - of, [3]

l‘iquation I3] represents the traditional concept of group
selection, and is represented in Fig. l by the broken line.
.\ny point to the right of ti.c.. above) this line increases the
fitness of the group.

Obviously, there is some overlap between l‘iqs. [2] and [3].

Proc. Nat. Acad. Sci. USA 72 (1.975)

that is, some traits selected for by individual selection also
increase the group’s ﬁtness. The problem of group selection is
to determine if and how those traits that are advantageous to
the group, yet outside the realm of individual selection, can
be selected for, and conversely, how those traits that are dis-
advantageous to the group, yet within the realm of individual
selection, can be blocked.

If the deme contains more than one trait-group (i.e., there is
a dispersal phase), the per capita ﬁtness changes of the A and
B types for the deme are respectively: I

231mm + No. - l/N)f.] ZNbilNaifr]
2 Na, and 2: Nb,

These are simply the weighted averages of individual ﬁtness
changes over all the trait-groups in the deme. Each trait-
group is assumed to have an equal overall density N and a,,b,
are the proportions of the A and B types in each trait-group i.

As before, the A-typc is selected for only if it has the highest
per capita fitness.

2 Na,[fd + Ma, — 1/N)f,] Z.\*b,[.\-'a,f,1

_ ..__ _ __. ..__ 4
Z: Ara! Z {Vbi [ ]
Z i‘iaifd Z A'ibiaifr Z: Narfat - l/ler
- 1 AA- __ > - t _ _ *1. >
2 Na, 2 bi

”—27——
I

2 “lb! 2 0:2
f.>/. N -'—~-—-— — ' +1 [51
‘ 2”:- Z a:
I I
Eq. [5] gives the condition for selection of the A-trait in the
dome. It is the. same as expression [2] with the exception of the

term:
2 aibi Z ”'12
v 7

.V " [6]

Z I): Z a:
I 1

The value of this term depends on the composition of the
trait-groups. (liven a single trait-group or trait-groups in
which the proportions of A and B-types are identical, term [6]
equals zero and liq. [5] reverts to Eq. {2], the conditions for
individual selection. If the types are completely segregated,
such that any trait-group consists either entirely of A or
entirely of B, then term [6] equals —N and liq. [5] reverts to
liq. [3], the conditions for pure group selection. Intermediate
variation in trait-groin) crmiposition yields intermediate
solutions. Thus, the effect of increasing the variation in the
composition of trait-groups is to push the system towards
group selection. Graphically this is represented by rotating
the solid line (giving the set of traits actually selected for) in
Fig. l counter-chickwise, until it is coincidental with the
broken line (arrow).

A variation greater than zero in the composition of trait-
groups will be met by any stochastic process. If the placement.

 

 

of i}'])t'.\ into the trait—groups is randomly determined, then
the variation in compositicm will follow the binomial distri-
bution. In this case the expected value of term [6] is always
—I regardless of trait-group size (N) or the overall frmplcncy
of the -l-typc in the deme to). liq. [5] then becomes:

f, > o. [7]

Proc. Nat. Acad. Sci. USA 72 (1.975)

In other words, given a random distribution of types into
trait-groups, any trait that increases the absolute ﬁtness of
the donor, regardless of its relative ﬁtness will be selected for.
Graphically, the solid line is rotated until it is coincidental
with the y-axis. I am very. grateful to F. M. Stewart for the
proof of this, which will be presented elsewhere (the proof is
not dependent on equal N in each trait-group).

If the variation in the composition of trait-groups is greater
than random, term [6] yields values of less than -1, and
altruistic traits that actually decrease the ﬁtness of the donor
can be selected for, such as alarm calls. This is also indepen-
dent of A’s frequency in the deme.

Kin that remain close to each other constitute one way of
generating this greater than random variation (kin selection
is thus a subset of this theory) but it is not the only way.
Animal distributions are often found to be “patchy” or with a
greater than random variation (9, 10). As one example, con-
sider a situation in which larval insects are deposited into the
trait-groups by adult females. The larvae upon hatching
intermix within the trait-group, and so do not fall under the
traditional concept of kin-selection. Assume that the females
enter the trait-groups at random, N to a trait-group, so that
as far as the female distributions of A and B types are con-
cerned, term [6] equals -1. Each female then lays e eggs.
Term [6] for the larval trait-group composition is now:

2 Gib! Z 0:2
eN '2 b - i a = e(—1) [8]
. i . i

t

 

i.e., the proportions remain the same but the density is raised
by a factor e, and the right hand side of Eq. [5] becomes
f,(1 — e), highly negative.

J. l\~Iaynard-Smith’s (11) model of group selection is rather
similar to the one presented here. He had the general concept
of trait-groups, but apparently thought it was still necessary
for altruistic traits to drift to ﬁxation in some of the trait-
groups for selection to occur. Genetic drift is not necessary for
this model.

DISCUSSION

The process of group selection postulated here can be visual-
ized in Fig. 2, showing two trait-groups with differing pro-
portions of A and B types (2A). The A-trait is an “altruistic”
defense behavior, such as a warning cry. While the animals
are in their trait-groups, predation occurs and within each
trait-group the B’s fare better than the A's. However, con-
sidering both groups together the opposite is true, that is, the A’s
fare better than the 8’8 (213). This is because due to the A-trait,
the trait-group with the most A’s has less overall predation
upon it.

Were the groups to remain in isolation this would mean
nothing, and the A’s would rapidly be eliminated. llowever,
all animals leave the trait—groups (2C), each has a single off-
spring, and the population settles back into the trait-groups
(2D). The increased proportionality of the .\-type for the en-
tire system is now realized within each trait-group, and by
this process B is eliminated from the system.

Notice that this form of group selection never really vio-
lates the concept of individual St'lH'llHl]. It is always the type
with the highest per capita fitness that. is chosen, but when the
effect of more than one trait-group is considered, these are. the
very types that behave altruistically.

A Theory of Group Selection 145

   

o =.75 b=.25 l
a=.solB:.50

l o'.25 b'.7$

   

o '.66 b=.33 I

 

aesslB-As
.
U
A.“A.‘. A.‘.A A.4
A: I‘A.4.A30,Ac
'4 I4 A

   

o . .30 b- .20 A
small-.45

Fin. 2. Illustration of the group selection process. See (art for
explanation.

The. extent to which this process of group selection occurs
depends on (1) the validity of the trait-group concept, and
given this, (2) the variation in the composition of trait-
groups. Both may be expected to vary widely among orga-
nisms, depending on behavior and habitat. In particular, small
insular habitats might constrict the deme to the size of a
single trait-group and push the system towards “individual"
selection; and spatial heterogeneity, by partitioning the deme
spatially, may be expected to enforce trait-groups and en-
hance group selection. In any case, the traditional concepts of
group and individual selection appear to be two extremes of a
continuum, with systems in nature operating in the interval
in between.

I am indebted to Conrad Istock, Ii). 0. Wilson, IS. IS. Werner,
l). J. Hall, (l. C. Williams, ll. Caswell, F. M. Stewart, B. Levin,
and the lieology group at Michigan State for helpful criticism
and discussion. This work was supported by National Science
Fonndaticm (irants (JBSSOSS and IRISH-20.3.30. This is con-
tribution no. 267, Kellog Biological Station.

1. \Vynne—l‘ldwards, V. (1‘. (1962') Animal Dispersion in. Rela-
tion to Social lfcluu'z'm' (Oliver and Boyd, l‘ldinlnn'gh, Scot-
land), 6.32} pp.

2. Haldane, .l. B. S. (192:2) 'I’hc ("rinses of Erolulz‘on (Longmans,
(irecn & (‘o.; reprinted in 1966 by Cornell University Press,
Ithaca. N.Y.; ('ornell paperbacks), 213.3 pp.

2-}. Wright. S. (lit-1.3) “'l‘einpo and mode in evolution: A critical
revievv," Is'coltmg/ 26, 413- 41‘).

146 Zoology:

Wilson
6

Levins, R. (1970) “Extinction,” in Some .llathematical
Questions in Biology. Lectures 0n .1latlicmatics in. the Life
Sciences, ed. Gerstenhaber, M. (American Mathematical
Society, Providence, R.I.), Vol. 2, pp. 77—107.

Boorman, S. A. & Levitt, P. R. (1973) “Group selection on
the boundary of a stable population,” Theor. Pop. Biol.
4, 85—128.

Levin, B. R. & Kilmer, W. L. (1974) “Interdemic selection
and the evolution of altruism: a computer simulation
study,” Evolution, in press.

9.

10.

ll.

Proc. Nat. Acad. Sci. USA 72 (1975)

Wilson, E. O. (1973) “Group selection and its significance
for Ecology,” BioScieiice 23, 631-638.

Johnson, C. G. (1969) Migration. and Dispersal of I "sects by
Flight (Methuen & Co., London), 763 pp.

Hutchinson, G. E. (1953) “The concept of pattern in
ecology,” Proc. Acad. Natural Sci., Philadelphia, 105, 1—12.
Lloyd, M. (1967) “Mean crowding,” J. Anim. Ecol. 36, l—
'30

Maynard-Smith, J. (1964) “Group selection and kin selec-
tion,” A’ature 201, 1145—1147.

CONCLUSIONS

What are the implications of this model, if true? The concept
of structured demes admits the routine evolution of weak altruism,
but at its strongest it becomes conceptually identical to kin
selection. In other words, it cannot explain the evolution of any
trait that could not also be explained by kin selection, but it does
enlarge the number of situations in which such traits evolve.

It is also possible that both kin selection and this model
are a stronger evolutionary force than is currently believed. In
fact, I believe that in focusing on the concept of altruism, the
whole group selection controversy has been misplaced. The funda-
mental question is "do populations maximize collective fitness"--
whether they do so through selfish or altruistic means is a secondary
consideration. Even if the structured deme process can only select
for weak altruism, its role in selecting for or against traits that
have neutral individual selection value may well increase its

potency.

BIBLIOGRAPHY

BIBLIOGRAPHY

Boorman, S. A. and P. R. Levitt (1973). Group Selection on the
boundary of a stable population. Theoret. Pop. Bio. 4(1):
85-128.

Gilpin, M. E. (1975). The theory of group selection in predator—
prey communities. Princeton University Press, 97 pp.

Haldane, J. B. S. (1932). The causes of evolution. Longmans, Green
& Co. (reprinted in 1966 by Cornell University Press, Cornell
paperbacks, 235 pp).

Hutchinson, G. E. (1953). The concept of pattern in ecology. Proc.
Acad. Natural Sci. 105:1-12.

Johnson, C. G. (1969). Migration and dispersal of insects by
flight. Methuen & Co., 763 pp.

Levin, B. R. and W. L. Kilmer (1974). Interdemic selection and the
evolution of altruism; a computer simulation study.
Evolution (in press).

Levins, R. (1970). Extinction. In M. Gerstenhaber, ed. Some
Mathematical questions in biology. Lectures on mathematics
in the life sciences. American Mathematical Society,
Providence, R.I. 2:77-107. '

Lloyd, M. (1967). Mean crowding. J. Anim. Ecol. 36:1-30.

Maynard-Smith, J. (1964). Group selection and kin selection.
Nature 201(4924):1l45-1147.

Wilson, E. O. (1973). Group selection and its significance for
Ecology. BioScience 23:631-38.

Wright, S. (1945). Tempo and mode in evolution: a critical review.
Ecology 26(4):415-19.

Wynne-Edwards, V. C. (1962). Animal Dispersion in Relation to
Social Behavior. Oliver & Boyd, 635 pp.

nICHIan STRTE UNIV. LIBRRRIES
\WlWIWlII‘“IWWWWIIWI’WNIHWl
31293101390874