Reimpression du
Reprinted from
Patterns of life-history diversification in North American
fishes: implications for population regulation
K. O. WINEMillER AND K. A. ROSE
1992
Number 10
Volume 49
.
.
Pages 2196-2218
(~nada
FISheries
and Oceans
~hes
et Oceans
~ in can.s. by The ~ ~ LIni8d
~ au C8n8da per The RU"98 Pr8SS ~
Patterns of Life-History Diversification in North American Fishes:
Implications for Population Regulation
Kirk O. Winemiller1 and Kenneth A. Rose
Environmental Sciences Division, Oak Ridge National Laboratory, Oak Ridge, TN 37831-6036. USA
Winemiller, K. 0., and K. A. Rose. 1992. Patterns of life-history diversification in North American fishes: impli-
cations for population regulation. Can.). Fish. Aquat. Sci. 49: 2196-2218.
Interspecific patterns of fish life histories were evaluated in relation to several theoretical models of life-history
evolution. Data were gathered for 216 North American fish species (57 families) to explore relationships among
variables and to ordinate species. Multivariate tests, performed on freshwater, marine, and combined data ma-
trices, repeatedly identified a gradient associating later-maturing fishes with higher fecundity, small eggs, and
few bouts of reproduction during a short spawning season and the opposite suite of traits with small fishes. A
second strong gradient indicated positive associations between parental care, egg size, and extended breeding
seasons. Phylogeny affected each variable, and some higher taxonomic groupings were associated with particular
life-history strategies. High-fecundity characteristics tended to be associated with large species ranges in the
marine environment. Age at maturation, adult growth rate, life span, and egg size positively correlated with
anadromy. Parental care was inversely correlated with median latitude. A trilateral continuum based on essential
trade-offs among three demographic variables predicts many of the correlations among life-history traits. This
framework has implications for predicting population responses to diverse natural and anthropogenic disturbances
and provides a basis for comparing responses of different species to the same disturbance.
Les caracteristiques du cycle biologique communes a plusieurs especes de poissons ont ete evaluees par rapport
a plusieurs modeles theoriques de I'evolution a I'interieur du cycle biologique. On a recueilli des donnees sur
216 especes nord-americaines de poissons (57 families) afin d'explorer les rapports entre differentes variables et
afin de classer les especes. Des tests multivaries, faits sur des matrices correspondant aux eaux douces, aux eaux
de mer et aux deux, ont regulierement fait ressortir un gradient qui associe les poissons a maturation lente a une
fertilite elevee, a la petitesse des oeufs et au nombre restreint de peri odes d'activite sexuelle au cours d'une breve
saison de fraie, et qui associe les traits opposes aux poissons de petite taille. Un deuxierne gradient marque a
indique des associations positives entre les soins des parents, la grosseur des oeufs et I'existence de saisons de
fraie prolongees. La phylogenie a des effets sur chacune des variables, et certains groupes taxonomiques supe-
rieurs sont associes a des strategies particulieres du cycle biologique. II tendait a exister un rapport entre la fertilite
elevee et I'aire de distribution des grosses especes en milieu marin. L 'Age a maturite, la vitesse de croissance des
adultes, la duree de vie et la grosseur des oeufs etaient tous en correlation positive avec I'anadromie. Les soins
des parents etaient en correlation inverse avec la latitude mediane. Un ensemble trilateral de donnees fondees
sur des compromis essentiels entre trois variables demographiques, permet de predire beaucoup de correlations
avec les caracteristiques du cycle biologique. Ce cadre d'examen est utile a la prevision des reponses de popu-
lations a differentes perturbations naturelles et d'origine anthropique, et il procure la base pour la comparaison
des reponses de differentes especes a une m~me perturbation.
Received October 29, 1991 Rec;u Ie 29 octobre 1991
Accepted May 8, 1992 Accepte Ie 8 mai 1992
(JB286)
in different geographical locations involving different fish
faunas.
Because it is qualitative and emphasizes the physiological
ecology of early life stages, the reproductive guild concept is
limited in its application to many practical problems. Much of
population biology and fisheries science is founded in mathe-
matical formulations, and a framework based on developmental
physiology does not easily yield quantitative predictions.
Hence, a general, yet quantitative comparative framework that
could interface with both qualitative schemes, like reproductive
guilds, and quantitative population models is desirable. To pro-
vide a further step toward a conceptual framework of fish eco-
logical strategies, we examine patterns of life-history variation
among North American fresh water and marine fishes and eval-
uate gradients of variation in relation to several earlier models
of life-history evolution.
Because life-history traits are also the fundamental deter-
minants of population performance, the investigation of life-
B alon (1975) listed the requirements for a comparative
framework useful for predicting the response of fish pop-
ulations to different kinds of environments and disturb-
ances. Such a framework should contain few categories and
allow researchers "to build from bits and pieces of available
information about reproductive strategies" (Balon 1975).
Moreover, it should group similar species irrespective of phy-
logenetic origin. In other words, adaptive convergences should
be stressed over phylogentic affiliations. Balon's (1975; Balon
et al. 1977) reproductive guild framework was based on the
premise that environmental requirements and adaptations of
early life stages are likely to account for a large amount of the
variance in densities and geographical distributions of fish pop-
ulations. Reproductive guilds permit researchers and resource
managers to identify common ecological features and problems
ICunent address: Department of Wildlife and Fisheries Science,
Texas A&M University, College Station, TX 77843-2258, USA.
CM. J. Fish. AqIIat. Sc;.. Vol. 49. 1992
2196
relation to different scales of variation in resources and sources
of mortality.
Materials and Methods
Life-History Data Set2
Estimates of fish life-history traits were obtained from
literature sources that summarize large amounts of quantitative
data for individual species (e.g. Carlander 1969, 1977; Hart
1973; and synopses of biological data published by Food and
Agriculture Organization, Rome). In some instances, we
consulted the original studies cited in the species synopses to
allow better judgement of the method of estimation and
reliability of data. Most fish species exhibit considerable
interdemic variation in life-history traits over their geographical
ranges. Therefore, we determined the average or modal value
of traits by using data from populations located near the center
of species' ranges. For example, if a freshwater species ranged
from central Canada to the Tennesse River, we sought studies
conducted near the latitudes of the Great Lakes. When limited
data were available near the center of the range, we sought
estimates from peripheral populations in an incremental fashion
(working outward from the center of the range). When no
reliable data were found for a given trait, that cell in the species
by life-history trait matrix was left blank and all calculations
calling for the trait eliminated the species from the analysis.
Whenever maturation and growth data were reported for the
sexes separately, we used the estimates for females. In some
instances, total lengths were calculated from standard lengths
or fork lengths using published conversion equations. Because
no conversion equations were available for North American
cavefishes (Amblyopsidae) , we estimated conversion ratios
from measurements of photographs.
Data were obtained for the following 16 life-history traits.
(I) Age at maturation - the mean age at maturation in years,
or when estimates were in summarized form, the modal age of
maturation.
(2) Length at maturation - the modal length at maturation
in millimetres total length (TL), or if not reported, either the
median or minimum length at maturation.
(3) Maximum length - the maximum length reported in
millimetres TL.
(4) Longevity - maximum age in years.
(5) Maximum clutch size - the largest batch fecundity
reported.
(6) Mean clutch size - the mean batch fecundity for a local
population, i.e.data from a specific location or ecosystem,
calculated as
"
L N/F;
(I) E = i-I
strategies is central to both theoretical ecology and resource
management. Life-history theory deals with constraints among
demographic variables and traits associated with reproduction
and the manner in which these constraints, or trade-offs, shape
strategies for dealing with different kinds of environments. Life-
history trade-offs may have a primarily physiological basis (e.g.
clutch size and investment per offspring; Smith and Fretwell
1974), a demographic basis (e.g. intrinsic rate of increase and
mean generation time; Birch 1948; Smith 1954), an ecological
basis (e.g. provision of parental care and clutch size; Sargent
et al. 1987; Nussbaum and Schultz 1989), or a phylogenetic
basis (Gotelli and Pyron 1991). Of course, organisms consist
of complex suites of life-history traits, so that genetic corre-
lations between coevolved traits exhibiting a strong trade-off
can indirectly result in correlations with other traits (Pease and
Bull 1988). For example, Roff (1981) reevaluated Murphy's
(1968) findings for clupeid life histories and concluded that
variation in reproductive life span could be explained by its
correlation with age at maturity, rather than as a direct evolu-
tionary response to variation in reproductive success.
Insights into the evolutionary response of life-history param-
eters to different environmental conditions and spatiotemporai
changes usually come from two approaches: theoretical models
of life-history evolution (e.g. Cole 1954; Cohen 1967;
Goodman 1974; Schaffer 1974; Green and Painter 1975; Boyce
1979; Roff 1984; Sibly and Calow 1985, 1986) and analyses
of empirical patterns (e.g. Kawasaki 1980; Stearns 1983; Dun-
ham and Miles 1985; Roff 1988; Winemiller 1989; Paine 1990).
Both of these approaches have relied, to a large degree, on the
r-K continuum (Pianka 1970) or similar unidimensional
schemes (e.g. bet-hedging (Murphy 1968) and iteroparity-
semelparity gradients (Cole 1954; Schaffer 1974» as the basis
for comparing alternative life-history strategies. Triangular
continua containing three endpoint strategies (r-, K -, and stress-
or adversity-resistance) have been adopted to interpret patterns
and consequences of observed life-history variation in plants
and insects (Grime 1977, 1979; Southwood 1977, 1988; Green-
slade 1983). Studying fishes in very different environments,
Kawasaki (1980, 1983), Baltz (1984), and Winemiller (1989;
Winemiller and Taphorn 1989) independently identified three
similar strategies as endpoints of a triangular continuum. Fol-
lowing Winemiller (1992), these can be classified as (I) small,
rapidly maturing, short lived fishes (opportunistic strategists),
(2) larger, highly fecund fishes with longer life spans (periodic
strategists), and (3) fishes of intermediate size that often exhibit
parental care and produce fewer but larger offspring (equilib-
rium strategists).
Here, we further evaluate the trilateral continuum model of
fish life-history strategies by analyzing data from 216 North
American freshwater and marine fish species. Even though
many North American species are not included here, the breadth
and evenness of phylogenetic coverage should be sufficient to
identify major axes of life-history variation and to ordinate spe-
cies into a framework of basic ecological and demographic
strategies. We adopt broad interspecific comparisons under the
assumption that consistent intercorrelations among life-history
features across widely divergent taxa are likely to reveal both
fundamental constraints and adaptive responses to environ-
mental conditions. Life-history traits and strategies are then
examined with respect to phylogeny, and observed patterns are
evaluated in reference to several models of life-history evolu-
tion and population regulation. Finally, we argue that greater
understanding of population regulation in fisheries can be
achieved by contrasting alternative life-history strategies in
Can. J. Fish. Aqua'. Sc;.. Vol. 49. 1992
.
:L. HI
where E is the mean clutch, Nj is the number of individuals in
age or size class i, Fj is the number of mature eggs per clutch,
and n is the number of age or size classes in the population.
(7) Egg size - the mean diameter of mature (fully yolked)
ovrian oocytes (to nearest 0.01 rom).
2 A database listing our principal literature sources. life history traits.
and numerical estimates is available. for a nominal fee. from the Dep-
ository of Unpublished Data. CISTI. National Research Council of
Canada. Ottawa. Ont. KIA OS2. Canada.
2197
where I, is the annual length gain for adults in age class i. Lj is
TL for an adult entering age class i. and n is the number of
adult age classes.
Within local populations and size classes. individual fish
exhibit considerable variation in life-history traits. Our analysis
assumes that measures of central tendency for populations near
the center of their species distributions allow the investigation
of relationships among life-history variables within a broad
inter-specific context.
Phylogeny and Ecological Data Set
Phylogeny of North American fishes follows Lee et al. ( 1980)
for freshwater fishes and Nelson (1984) for marine fishes. We
coded family and order as categorical variables for statistical
analyses of phylogenetic influences on life-history traits. Each
species was classified as either freshwater or marine depending
on where the greatest fraction of the life cycle occurred. Because
the majority of the estuarine populations extend into other
coastal marine habitats, estuarine fishes were included in the
marine category. We obtained data for marine populations of
Oncorhynchus mykiss (steelhead), Menidia beryllina (inland
silverside), and Gasterosteus aculeatus (threespine stickle-
back) and data for a landlocked freshwater population of
Oncorhynchus nerka kennerlyi (kokanee salmon).
For each species, the ecological data set consisted of varia-
bles characterizing its geographical range, general habitat, and
general ecological niche. We recorded the midrange latitude
and total range in latitude for each species based on range maps
or verbal accounts of species' ranges. Each species was clas-
sified as either benthic (I), epibenthic (2), or pelagic (3) based
on accounts of the normal depth distribution of adult fishes.
The basic adult habitat was classified as either caves or springs
(0), small cold-water streams (1), small warmwater streams
(2), river channels (3), river backwaters and lakes (4), estuaries
(5), marine benthic (6), or marine pelagic (7). Fishes that are
common in two or more categories or occupy intermediate hab-
itats were assigned fractional values (e.g. habitat = 2.5 for
/ctiobus bubalus (smallmouth buffalo) which commonly occurs
in both rivers and lakes). Based on summarized diet informa-
tion for adults, trophic status was classified as either detritivore/
herbivore (1), omnivore (2), invertebrate-feeder (3), or pisci-
vore (4). Fishes that consume large quantities of both inverte-
brates and fishes formed a fifth intermediate category (e.g.
trophic status = 3.5 for Micropterus dolomieu (smallmouth
bass». The relative migratory behavior of each species was
classified under the headings anadromous (I), sedentary (0),
and catadromous (-1). Fishes exhibiting spawning runs from
lakes into rivers or from rivers into affluent tributaries (pota-
modromy) were classified as 0.5, and fishes exhibiting spawn-
ing migrations from nearshore to offshore were classified as
-0.5.
Data Analysis and Comparisons
Analyses based on broad interspecific comparisons yield pat-
terns that have resulted from many generations and thousands
of years of evolution. As a consequence, comparisons involv-
ing many taxa and large phylogenetic breadth should contain
fewer idiosyncracies due to genetic correlations carried along
within a particular phylogenetic lineage (= phylogenetic con-
straints). If divergent lineages are not given fairly equivalent
representation, taxonomic bias can enter into interspecific com-
parisons (Pagel and Harvey 1988). Some have even argued
against the use of species in comparisons (e.g., Ridly 1989).
(8) Range of egg sizes - the range of diameters for mature
ovarian oocytes reported for a local population (0.01 mm).
(9) Duration of spawning season - number of days that
spawning or early larvae were reported.
(10) Number of spawning bouts per year - the mean number
of times an individual female was reported to spawn during a
year. Because multiple spawning bouts are difficult to document
in wild fishes, many of the estimates used in this analysis are
probably underestimates. Several recent studies of small
cyprinids and percids indicate that multiple clutches may be
common in small fishes having small clutches (e.g. Heins and
Rabito 1986; Heins and Baker 1988; James et al. 1991). When
two fairly distinctive size classes of ova were reported in mature
ovaries and other evidence was consistent with a hypothesis of
repeat spawning, we recorded the species as having two bouts
per year. Following Hubbs (1985), we used an average
interbrood interval of 10 d for estimates of spawning bouts for
several darters (Etheostoma, Percidae) that exhibited strong
evidence of multiple clutches.
(11) Parental care - following Winemiller (1989), quan-
tified as I.x; for i = I to 3 (XI = 0 if no special placement of
zygotes, 1 if zygotes are placed in a special habitat (e.g.
scattered on vegetation, or buried in gravel), and 2 if both
zygotes and larvae are maintained in a nest; X2 = 0 if no parental
protection of zygotes or larvae, I if a brief period of protection
by one sex « 1 mo), 2 if a long period of protection by one
sex (> I mo) or brief care by both sexes, and 4 if lengthy
protection by both sexes; and X3 = 0 if no nutritional
contribution to larvae (yolk sac material is not considered here),
2 if brief period of nutritional contribution to larvae (= brief
gestation « I mo) with nutritional contribution in viviparous
forms), 4 if long period of nutritional contribution to larvae or
embryos (= long gestation (1-2 mo) with nutritional
contribution), or 8 if extremely long gestation (>2 mo). We
reason that, in terms of benefits received by offspring, gestation
with nutritional contribution is approximately equivalent to
biparental brood guarding during an equivalent time period.
Parental care values (I.xJ ranged between 0 (no care) and 8
(long gestation in some embiotocid fishes) in the North
American fish data set.
(12) Time to hatch - the mean time to hatch within the range
of values for average midseason temperatures, or when not
reported, the mean, modal, or midrange time to hatch at the
highest temperature reported within a reasonable range (relative
to local ambient temperatures) for a given locality.
(13) Larval growth rate - mean increment in millimetres TL
during the first month following hatching. We subtracted the
length of larvae at hatching from the mean length attained after
the first month. For the few species with larvaJ stage duration
<1 mo, we converted mean daily growth rates to mean
increments for 30 d.
(14) Young of the year (YOY) growth rate - mean increment
in millimetres TL during the first year following hatching or
independent life for viviparous fishes. We subtracted the length
of larvae at hatching from the mean or modal length attained
after the first growing season.
(15) Adult growth rate - mean increment in millimetres TL
per year of life over an average adult life span (in this case,
data were not weighted by sample size of age cohorts).
(16) Fractional adult growth - mean fraction of millimetres
TL gained per year in a normal adult life span, calculated from
II
L ljlLj-J
G = ;-2
n
Can. J. Fish. Aqua/. Sci., Vol. 49, /992
2198
only one species per genus. The species used to represent gen-
era were selected in alphabetical order to reduce experimenter
bias.
To further test relationships between life-history patterns and
the species' environmental biology, we performed canonical
discriminant function (COP) based on nine life-history varia-
bles and three ecological variables (habitat, trophic status,
migration) and parental care recorded as classification varia-
bles. CDF derives canonical variables from the set of life-his-
tory variables in a manner that maximizes multiple correlations
of the original variables within groups. To show general asso-
ciations between ecological groupings and suites of life-history
traits, we plotted the means and standard deviations of each
class on the first two COF axes.
Results
Univariate Comparisons
Life-history parameters showed large variation both within
the entire data matrix and within orders containing the largest
number of species. Standard deviations approached, and in
many instances exceeded, the magnitude of the mean values
for life-history traits (Table 1). The pygmy sunfish, Elassoma
zonatum (Centrarchidae), was the smallest fish in the overall
North American data set (minimum size at maturation
= 25.0 mrn). The largest fishes were the Atlantic sturgeon,
Acipenser oxyrhynchus (Acipenseridae) (length at maturation
= 2.5 m), Pacific halibut, Hippoglossus stenolepis
(Pleuronectidae) (maximum length = 2.7 m), and ocean
sunfish, Mola mola (Molidae) (maximum length = 3 m). The
smallest maximum clutches (batch fecundities) were recorded
for two live bearing surfperches (Embiotocidae) ,
Hyperprosopon argenteum (12) and Cymatogaster aggregata
(20). The largest clutch size estimates were for the ocean sunfish
(average = 300 x 1(('), Atlantic cod, Gadus morhua (Gadidae)
(maximum = 12 x 1(('), and ~n, Megalops atlanticus
(Elopidae) (maximum = 12.2 x 1()6). Reported estimates of
average egg sizes ranged from a minimum of 0.45 mrn in
diameter for the bay anchovy, Anchoa m;tchilli (Engraulidae),
to a maximum of 20.5 mrn for the mouthbrooding gafftopsail
catfish, Bagre marinus (Ariidae). Average larval growth rates
ranged from a minimum estimate of 1.3 mrn TUmo for lake
whitefish, Coregonus clupealormis (Salmonidae), to a high of
69.9 mm TUmo for longnose gar, Lepisosteus osseus
(Lepisosteidae) .
Comparisons between marine and freshwater fishes showed
differences in the distributions of life-history attributes (notable
examples illustrated in Fig. 1). Statistically significant mean
differences are reported with the following notation: t = value
from two-sample I-test, z = value from Mann-Whitney U-test.
Because both data sets were biased somewhat in favor of larger,
commercial species, our interpretations of these univariate
comparisons are tentative. As a group, marine fishes matured
later (mean marine (m) = 3.38 yr, mean freshwater (/)
= 2.74 yr, t = 1.97, df = 194, P < 0.05; Fig. la), matured
at larger sizes (m = 320 mrn, 1= 186 mrn, t = 4.03, df =
202, P< O.(xx)l), lived longer (m = 13.0 yr,l = 9.7 yr, t =
2.19, df = 185, P < 0.05), had larger mean clutches (m =
1 554 400, 1= 113376, t = 4.34, df = 189, P < 0.(xx)1;
Fig. Ib), had longer spawning seasons (m = 103 d,l = 59 d,
t = 5.55, df = 213, P < 0.<XX>1; Fig. lc), had larger YOY
growthrates(m = 131.2mm/yr,j= 98.4mm/yr,z = 2.10,
P < 0.0025), and had larger adult growth rates (m = 50.8 rnm/
yr,1 = 30.5 rnm/yr, z = 3.12, P < 0.001) than freshwater
In essence, the comparative approach requires that a pattern be
repeated consistently within a variety of taxa (i.e. conservation
or convergence of pattern), or the effect of phylogeny be held
constant (or adjusted for statistically), if hypotheses of adap-
tation are to be tested. To maximize the likelihood that emer-
gent patterns reflect adaption, we examined life-history patterns
based on a variety of different combinations of life-history traits
and various ecological and phylogenetic subsets of the overall
data set.
Because some distributions of raw life-history traits were log-
nonnal, data were In-transformed for parametric statistical tests.
All statistics were calculated using SAS (SAS Institute Inc.
1987). Univariate comparisons between various subsets of the
data set used either I-tests (two-sample, two-tailed) or Mann-
Whitney U-tests when data were interval or did not approximate
a normal distribution. Bivariate relationships among four cat-
egorical ecological variables and parental care (ordinal scale)
were also analyzed using Spearman's rank correlation coeffi-
cient. Chi-square tests for goodness of fit were used to compare
frequency distributions of life-history traits for marine and
freshwater fishes.
Nested analysis of covariance was used to test for effects of
phylogeny and body size on life-history traits. Order and family
were used as independent variables in tests of phylogenetic
effects (family nested within order). Mean TL at maturation
(In-transformed) was used as an index of body size. For addi-
tional insights into the effects of phylogenetic affiliation on life-
history patterns, bivariate relationships among several key life-
history variables were analyzed within several orders that con-
tained many species.
To explore patterns of association among life-history traits
and ordination of species, a series of principal components
analyses (PCA) was performed on In-transformed life-history
data. All PCAs were calculated from the correlation matrix to
standardize for the influence of unequal variances. Because data
were lacking for some traits for some species, analyses that
involved more life-history variables resulted in ordination of
fewer species. Therefore, separate PCAs were performed on
marine, freshwater, and combined fish data sets: one using 12
relatively nonredundant life-history variables (analyses omit-
ting maximum length (redundant with size at maturity), max-
imum clutch (redundant with average clutch), range of egg size
(redundant with egg size), and either length or age at maturation
(highly correlated with each other)) and another using only five
life-history variables (length at maturation, mean clutch, egg
size, spawning bouts, parental care). The five life-history var-
iables retained for the five-variable analyses were selected based
on their dominant influence in the 12-variable PCA models,
except that length at maturation was substituted for age at matu-
rity in the five-variable data set in order to increase the number
of species retained in the analysis.
Although values for parental care were ordinal, we included
In-transformed parental care values in multivariate analyses
because they approximated a normal distribution within most
groupings and were derived from an algorithm that combined
several independent attributes into a single numeric value.
Because some life-history researchers have identified an influ-
ence of body size on other life-history traits, some PCAs were
done both with and without length partialled-out of the other
variables (i.e. multivariate analyses are based on relationships
among the residuals from regressions of variables with length).
To test for potential biases resulting from disproportionate
inclusion of some taxa in the global data set (e.g. Lepom;s spe-
cies. N = 8), PCA was also performed on a data set containing
2199
Call. J. Fish. Aquat. Sci.. Vol. 49. 1992
