Professional Documents
Culture Documents
Allometry and Sexual Selection of Male
Allometry and Sexual Selection of Male
doi:10.1093/beheco/arh141
Advance Access publication 18 August 2004
Both male and female Wellington tree weta, Hemideina crassidens, use cavities in trees as diurnal shelters. That these galleries are
often limiting in nature offers males the opportunity to increase their reproductive success by monopolizing galleries and the
females residing in them. Male H. crassidens, can mature at either the 8th, 9th, or 10th instar, whereas females mature at the 10th
instar only, and male head (and mandible) size positively covaries with ultimate instar number. It has been suggested that males
fight for control of galleries by using their enlarged mandibles as weapons, and males with larger mandibles control galleries with
more females. In the present study, I present a statistical examination of sexual dimorphism, showing that traits related to head
size are on average significantly larger in males, whereas traits related to body size are on average significantly larger in females. I
tested three predictions addressing the hypothesis that sexual selection is driving megacephaly in male H. crassidens. First, as
predicted, traits related to head size show a positive allometric relationship with body size in males but not in females. Second,
adapting a novel statistical technique based on maximum likelihood and bootstrapping revealed that males, but not females,
exhibit a multimodal distribution in head and body size traits. This is likely a consequence of males maturing at one of three
instars, which results in positive covariance between the ultimate instar number and morphological traits. Third, as predicted,
single adult males with larger heads reside in galleries housing larger groups of adult females. Key words: allometry, harem success,
sexual selection, tree weta, weaponry. [Behav Ecol]
ales in many animal species control resources required with rivals (Baker and Wilkinson, 2001; Eberhard, 2002;
M for breeding by sexually receptive females or by compet-
ing directly for control of harems (Emlen and Oring, 1977).
Eberhard and Gutiérrez, 1991; Forsyth and Alcock, 1990;
Gould, 1974; Hanley, 2001; Palestrini et al., 2000; Simmons and
This situation creates intense intrasexual selection among Tomkins, 1996; Tatsuta et al., 2001). Although the reason why
males and can lead to selection for behaviors (Andersson, sexually selected traits tend to exhibit positive allometry is not
1994) or traits (e.g., elongate mandibles, antlers, and horns of certain, Baker and Wilkinson (2001) suggested that it is because
particular beetles and flies; for review, see Dodson, 2000; smaller-bodied individuals have relatively little to gain from
Emlen and Nijhout, 2000) that enable males to be more com- high investment in such traits. On the other hand, Petrie (1988)
petitive in contests for resources or harems. suggested that only traits used to signal competitive ability
One result of intense competition among males for mates is to conspecifics should exhibit positive allometry, and there is
selection for alternative reproductive strategies and tactics in no reason to expect allometric slopes greater than one in
which dominant males exhibit one phenotype and sub- structures that function only as weapons. Empirical evidence
ordinate males exhibit another (for review, see Brockmann, appears to contradict Petrie’s (1988) suggestion, however, as
2001). The production of two or more distinct phenotypes Eberhard and colleagues (Eberhard and Gutiérrez, 1991;
can result in a discontinuous or multimodal distribution of Eberhard et al., 1998) have demonstrated positive allometry
characters such as weapon size (Brockmann, 2001; Emlen and in structures in beetles (horns) and earwigs (cerci) that appar-
Nijhout, 2000). The evolution of two different alternative ently serve strictly as weapons, although a signaling function for
strategies, each with its own morphology and behavior, can these devices has not been ruled out (Eberhard, 2002).
evolve and is common (Shuster and Wade, 2003); however, The net effect of sexual selection, therefore, could be the
the evolution of three or more alternative phenotypes is uncom- evolution of alternative reproductive phenotypes within a sex
mon (Sinervo and Lively, 1997; Zimmerer and Kallman, with each phenotype expressing differently a trait that is
1989; but see Shuster, 1987). positively allometric and unimodal in distribution.
Characters under sexual selection often show allometric Hemideina tree weta (Orthoptera: Tettigonioidea: Anosto-
values greater than 1.0 when scaled on body size; that is, larger stomatidae) are a group of large, flightless, nocturnal insects
individuals have disproportionately larger trait values than do endemic to New Zealand (Gibbs, 2001). Anostostomatids
smaller individuals of a species (for review, see Andersson, 1994; exhibit various forms of male weaponry, including elephan-
Emlen and Nijhout, 2000; but see Eberhard, 2002). For tine tusks (e.g., Motuweta isolata, Gibbs, 2001) and enlarged
example, positive allometry is exhibited by traits used by mandibles (e.g., Hemideina tree weta and Anostostoma king
females in choosing a male mate (Alatalo et al., 1988; Green, crickets, Field and Deans, 2001; Gwynne and Jamieson, 1998;
1992; Petrie, 1988, 1992) and traits used by males in combat see also Hudson, 1920; Koning and Jamieson, 2001; Moller,
1985; Spencer, 1995).
In the sexually dimorphic Wellington tree weta, Hemideina
Address correspondence to C.D. Kelly. E-mail: cdkelly@utm. crassidens, males use their enlarged heads and mandibles in
utoronto.ca. fights for possession of shelters in tree cavities (Kelly CD,
Received 2 October 2003; revised 14 May 2004; accepted 1 June unpublished data). Within these cavities, known as galleries,
2004. one to several females refuge during the day (Field and
Behavioral Ecology vol. 16 no. 1 International Society for Behavioral Ecology 2005; all rights reserved.
146 Behavioral Ecology
METHODS
This study was conducted February 2001, March–May 2002, and
March–April 2003 on Te Hoiere/Maud Island, New Zealand
(41 029 S, 173 549 E), a 309-ha scientific reserve free of alien
predators (e.g., rodents [Mus and Rattus spp.] and stoats
in which Y is the log of head length or width, X is the log of All correlations were significant at the appropriate p value (male
pcorrected ¼ .007 and female pcorrected ¼ .005) after Bonferroni
femur length, bi is the regression coefficients, and e is the correction.
random component with assumed normal distribution, mean
zero, and common variance. If b2 does not differ significantly
from zero, then the relationship between head size and body Chapman, 2001), which was developed to evaluate hypotheses
size shows no significant deviation from linearity, and that multiple normal distributions comprised trait distribu-
consequently, determination of the body size switchpoint is tions (e.g., different arthropod instars). This statistical
not justified (Eberhard and Gutiérrez, 1991). approach provides an objective comparison of various
Eberhard et al. (2000) and Kotiaho and Tomkins (2001) plausible hypotheses of grouping when faced with a data set
suggest modifying Expression 1 to discriminate morphs based suspected of comprising more than one subdistribution.
on the dimorphic character itself rather on some correlate Briefly, MMA achieves this by (1) choosing the model with
of it, most commonly, body size. To accomplish this, they the number of groups that best fits the data (i.e. the best-
recommend simply substituting the Y with X and X with Y supported log-likelihood score for the hypothesized number
to get of groups) and (2) estimating the parameters (means,
Y ¼ b0 þ b1 X þ b2 Y 2 þ e; ð2Þ variances, and mixing proportions) of each of the subdis-
tributions for the best-fitting model (Hunt and Chapman,
where the terms are as defined in Expression 1. 2001). Next, MMA uses parametric bootstrapping to compare
If the relationship between male head size and body size is hypotheses that different numbers of groups comprise the
discontinuous and a switchpoint exists, I tested whether data. The power of a bootstrap test was calculated when the
additional switchpoints were present as outlined in Eberhard bootstrap failed to reject the null hypothesis (Hunt and
et al. (2000). If males exhibit a trimorphsim in head size, then Chapman, 2001).
it is possible that two switchpoints exist (i.e., a separate slope All statistical analyses were performed by using JMP 5.0
for each instar or morphotype). (SAS Institute, 2002) for Mac OSX and MMA software pack-
It is possible for a trait to be multimodally distributed and age for MS-DOS available at www.paleodb.org/paleosource
exhibit a linear relationship with body size if each mode (see also Hunt and Chapman, 2001).
exhibits continuous variation in trait size. Therefore, males
could have alternative strategies based on morphotype
without exhibiting the clearly demarcated switchpoints seen RESULTS
in beetles and earwigs (see Eberhard and Gutiérrez, 1991).
Male and female morphology
Detecting multimodality is important to this study because
adult male H. crassidens are known to exhibit trimodality in A total of 110 (59 males and 51 females), 325 (143 males and
the laboratory associated with variation in instar number. I am 182 females), and 317 (159 males and 158 females) adult tree
testing the prediction that trimodality exists in the wild as weta were collected and measured in 2001, 2002, and 2003,
well. Therefore, in cases in which I found a linear relation- respectively. There were no significant within-sex differences
ship, I identified whether trait distributions comprised among years for any measured morphological trait (one-way
a mixture of normal distributions in both males and females ANOVA, all tests p . .05). Therefore, data for the 3 years were
in two separate steps. First, I visually inspected normal pooled for each sex.
quantile plots to determine whether each trait was normally There was male-biased sexual dimorphism in the four traits
distributed. Normal quantile plots are recommended (Sokal related to head size and female-biased dimorphism in the
and Rohlf, 1995) because other methods have either low three traits related to body size (Table 2). Males showed more
power (e.g., Kolmogorov-Smironov) or perform poorly when variation than did females in all measured traits as reflected in
there are tied data (e.g., Shapiro-Wilks; Zar, 1999). If the trait relatively high coefficients of variation (CVs) (Table 2). As in
was not normal, I used maximum-likelihood analysis of other species of Hemideina (Field and Deans, 2001; Gwynne
mixture models (mixture models analysis [MMA], Hunt and and Jamieson, 1998), there was directional asymmetry in
148 Behavioral Ecology
Table 2
Sample size, mean, and CV for absolute values of head size (mm) and body size (mm) in male and female Hemideina crassidens
Males Females
Dimorphism
Trait Mean 6 SE Range CV (%) n Mean 6 SE Range CV (%) n (% difference)
Weight 3.95 6 0.07 1.757.55 32.08 332 4.75 6 0.05 2.508.20 21.11 368 þ16.84
Left mandible length 10.18 6 0.14 5.2716.87 25.64 353 5.19 6 0.08 4.336.59 9.72 42 49.02
Right mandible length 9.24 6 0.12 4.9214.82 24.80 360 4.79 6 0.06 4.085.49 7.78 42 48.16
Values are mean 6 SEM or n. Differences between the sexes in each trait were tested by using a Student’s t test or Mann-Whitney U test
(denoted by *). Positive values for dimorphism denote female-biased difference and negative values denote male-biased difference. All
tests p , .001.
mandible length, with the left mandible always being longer length b2 ¼ 0.06 6 0.09, t ¼ 0.69, df ¼ 354, p ¼ .49; log10 head
than the right in both sexes. width b2 ¼ 0.09 6 0.13, t ¼ 0.68, df ¼ 356, p ¼ .50). Given
Both head size traits were positively correlated with body that the slopes (b2) did not differ from zero in any case,
size in all 3 years for each sex (Tables 1 and 3). Because the all relationships were linear and did not warrant further anal-
slopes of the relationship between each head size trait and ysis of body size switchpoints. A continuous relationship is
femur length did not differ among years for each sex, the data expected, despite the possibility of a trimodal distribution, if
were pooled by trait and by sex. The relationship between at least one individual is present in each trait-value bin and
head length and femur length was positively allometric in there are no clear breaks in the trait distribution.
males, as the slope was significantly greater than two (Figure 2 As predicted, inspection of normal quantile plots for data
and Table 3). In females, however, the slope did not differ pooled for 3 years showed that female head length, head
from one and was therefore isometric (Figure 2 and Table 3). width, and femur length were normally distributed (Figure
The relationship between head width and femur length was 3a–c) whereas male head length, head width, and femur
also positively allometric in males (slope . 1.0) but not in length were clearly not normal (Figure 3d–f). Male head
females (Figure 2 and Table 3). Males had significantly length was not normal in each of the three sampling years
steeper slopes than did females for each trait (test for (Figure 4), suggesting that the nonnormal distribution of the
heterogeneity of slopes: head length, T ¼ 9.78, df ¼ 354.50, pooled data is not an artefact of biased sampling for a
p , .0001; head width, T ¼ 8.35, df ¼ 355.34, p , .0001). particular head size in any one year. That is, all traits mea-
The relationship between head size and body size did sured in both males and females exhibited similar distribution
not differ significantly from linearity (log10 head length b2 ¼ patterns among years.
0.99 6 0.97, t ¼ 1.02, df ¼ 354, p ¼ .31; log10 head width b2 ¼ MMA suggests that there is a significantly higher likelihood
1.17 6 0.70, t ¼ 1.66, df ¼ 356, p ¼ .10 ), nor did the rela- that the distribution for each male trait is bimodal rather than
tionship between body size and head size (model 2; log10 head trimodal (Table 4). I am confident that a two-group model
Table 3
Slope, SE of slope, and the intercept of the allometric relationship between log head length and log femur length for male and
female Hemideina crassidens
Head length
2001 female 49 1.21 0.11 0.47 0.92 1.60 0.54 ns
male 59 2.47 0.12 1.82 2.21 2.78 0.84 ,.05
2002 female 174 1.08 0.05 0.30 0.94 1.22 0.67 ns
male 141 2.43 0.06 1.76 2.30 2.58 0.89 ,.05
2003 female 158 1.07 0.05 0.30 0.94 1.22 0.61 ns
male 157 2.44 0.06 1.77 2.32 2.57 0.90 ,.05
pooled female 381 1.09 0.03 0.32 1.00 1.17 0.63 ns
male 357 2.44 0.04 1.78 2.36 2.54 0.89 ,.05
Head width
2001 female 49 1.05 0.09 0.46 0.85 1.31 0.65 ns
male 59 1.86 0.09 1.33 1.68 2.08 0.86 ,.05
2002 female 174 1.00 0.04 0.39 0.92 1.09 0.76 ns
male 141 1.88 0.05 1.36 1.79 1.99 0.91 ,.05
2003 female 158 0.99 0.05 0.38 0.88 1.11 0.65 ns
male 159 1.84 0.04 1.31 1.75 1.94 0.91 ,.05
Pooled female 381 1.00 0.03 0.39 0.94 1.07 0.71 ns
male 357 1.86 0.04 1.33 1.80 1.93 0.90 ,.05
better explains the distributions of both head length and clearly demarcated (see Hunt and Chapman, 2001). That
femur length (body size) better than does a one-group model. each mode overlaps another also explains why the regressions
However, I am more cautious about choosing between two between head size traits and body size were linear and
and three groups for femur length because this test was nearly continuous (see above) as opposed to having a clear break in
significant at the 0.05 a-level (p ¼ .072) (Table 4) and because the distribution and, thus, a discontinuous relationship (e.g.,
tests for head length and femur length each suffered from low beetle horns; Eberhard and Gutiérrez, 1991). When taken
statistical power (0.19 and 0.17, respectively) (Table 4). Head together, regression analysis and MMA clearly show that head
width trimodality was the only hypothesis rejected with high traits are multimodally distributed but have a linear relation-
power (Table 4). Low power suggests that even if the best- ship with body size.
supported three-group model were true, data such as these
would be able to reject the two-group model only about 19%
Harem size and male head size
and 17% of the time (Table 4). That the mean head lengths
calculated under a three-group hypothesis (Table 5) fall My surveys of cavities revealed up to seven adult females
either within (intermediate ¼ 21–22 mm and large ¼ 25–26 residing in the same cavity with a single adult male (Figure 5).
mm) or close (small ¼ 15–16 mm) to each putative mode Modal harem size was zero (range ¼ 0–5, n ¼ 29) in natural
identified through visual examination provides some support cavities and 1.0 (range ¼ 0–7, n ¼ 79) in artificial cavities (both
for a trimodal distribution. Assuming a three-group distribu- years pooled). Resident male head length was significantly
tion in my data, the subdistributions show considerable positively correlated with the number of females in an artificial
overlap (Figure 3); a result consistent with Spencer’s (1995: cavity in 2003 (rs ¼ .34, n ¼ 55, p ¼ .011) but not in 2002 (rs ¼ .27,
see also Figure 1) laboratory data (because Spencer’s raw data n ¼ 24, p ¼ .21). Male head length was significantly positively
were unavailable to me, I could not subject them to MMA). correlated with harem size in both natural galleries (rs ¼ .52, n ¼
This overlap explains the low statistical power of the analysis 29, p ¼ .004) and in artificial galleries (pooled data for 2002 and
because MMA has higher resolving power when modes are 2003, rs ¼ .33, n ¼ 79, p ¼ .003). Pooling data for both cavity
Figure 3
Frequency distribution of head length, head width, and mean femur length in males (bottom) and females (top) for all years combined.
150 Behavioral Ecology
Table 4
Results of the maximum likelihood analysis for the one- through three-group models for each trait in male Hemideina crassidens
Trait Bootstrap test (H1 vs. H2) Observed 2l Bootstrap results Bootstrap p Power
Bootstrap results are the number of bootstrap replicates that exceeded the observed value for the log-likelihood ratio (2l), out of the total
number of replicates performed. An asterisk indicates that the model with fewer groups (H1) is rejected in favor of the model with more
groups (H2) at the a ¼ 0.05 level.
Kelly • Sexual selection on weaponry of male tree weta 151
Table 5
Mean (mm) and variance for the two- and three-group solutions for each trait in male Hemideina
crassidens
Head length 17.20 (4.14) 25.06 (6.62) 17.24 (4.21) 21.98 (0.001) 25.33 (5.83)
Head width 9.14 (0.63) 12.13 (1.41) 9.20 (0.68) 12.19 (1.20) 14.63 (0.00)
irreversible strategy (Brockmann, 2001) but simply males in The allometric values reported here for male head length
poorer condition (see below) ‘‘making the best of a bad job’’ are similar to values reported for weaponry in other insect
(Dawkins, 1980), thus maturing earlier and using a condi- species. For example, in their comparative analysis of 42
tional strategy based on the prevailing environmental condi- earwig species, Simmons and Tomkins (1996) found that
tions and their own phenotype (Brockmann, 2001). Whether males in 11 species showed positive allometry in forceps
the phenotype of a male tree weta is determined by either length (range of model I slopes ¼ 1.47–3.50), a trait used by
a genetic polymorphism or condition dependence remains to males in aggressive interactions with rival males (Briceño and
be investigated. Eberhard, 1995). Adult male staphylinid beetles, Leistotrophus
Male H. crassidens exhibited a positively allometric relation- versicolor, use their enlarged mandibles in combat with
ship between each head size trait and body size, whereas the conspecific males over access to dung and carrion, which is
relationship in females was isometric for those same traits. In used by sexually receptive females as a foraging site (Forsyth
both sexes, head length exhibited a steeper slope than did head and Alcock, 1990). Forsyth and Alcock (1990) reported
width, suggesting that length may be more important than is allometric values of 2.1 and 1.8 for the relationship between
width in male-male combat. That all relationships were also mandible length and head width and mandible length and
linear suggests that males of intermediate body size produce wing length, respectively. A sexual selection hypothesis is
intermediate head sizes compared with species that exhibit supported in that species for the evolution of enlarged
a sigmoidal or discontinuous scaling relationship (Emlen and mandibles because males with larger mandibles are better
Nijhout, 2000). Males of species in the latter category have fighters and mate with more females (Forsyth and Alcock,
either minimal or complete expression of the trait over a narrow 1990). Cephalic horn length in Onthophagus beetles is
range of intermediate body sizes (Emlen and Nijhout, 2000). considered to be important in male-male combat (Eberhard,
In the congeneric H. maori, Jamieson and colleagues 1982, 1979; Emlen, 1994), and allometric values of 7.5
(Gwynne and Jamieson, 1998; Koning and Jamieson, 2001) (Palestrini et al., 2000) and 6.23 (Eberhard et al., 1998) in
have found conflicting evidence of allometry in cephalic O. taurus and O. incensus, respectively, support the notion that
weaponry. Gwynne and Jamieson (1998) did not find a slope sexual selection is operating on this trait. Contrary to these
significantly greater than 1.0 either between mandible length examples, Eberhard reported that the front tibia in the male
and tibia length (slope ¼ 0.76) or between head width and tibia sepsid fly, Palaeosepsis dentatiformis, function as weapons yet do
length (slope ¼ 0.76). In contrast, Koning and Jamieson (2001) not exhibit allometric values significantly greater than 1.0
found positive allometry and a linear scaling relationship with (slope from OLS regression ¼ 0.96) and are only slightly
femur length in male head length (slope ¼ 1.31) and width sexually dimorphic. He suggested that because the front tibia
(slope ¼ 1.26). These differences may be the result of each study also serve a highly specialized function in courtship, their
focusing on different or inappropriate characters; perhaps head further specialization as weaponry is compromised. Similar to
width, as in H. crassidens (the present study), is not as important that of P. dentatiformis, the front femora of males in two species
in combat as is length, and femur length may be a better of Diptera (Chymomyza mycopelates and C. exophthalma) act as
indicator of body size than is tibia length. Given that slopes were weapons but exhibit isometric relationships with body size
inappropriately calculated by using model I least-squares (slopes from OLS regressions ¼ 0.919 and 0.745, respectively;
regression in both studies, a direct comparison of scaling Eberhard, 2002). These structures do not serve in courtship
relationships between H. crassidens and H. maori is not possible. or mating and are apparently not precluded from specializa-
tion as weapons (Eberhard, 2002). Eberhard (2002) suggested
that perhaps among species not all sexually selected traits will
have allometric values significantly greater than 1.0 if the costs
of building and carrying such a structure outweigh the
benefits. Controlled laboratory studies are required to better
understand how head traits scale with body size for each adult
instar in male H. crassidens and to estimate the fitness benefits
of maturing at different instars.
REFERENCES Gould SJ, 1974. The origin and function of ‘‘bizarre’’ structures:
antler size and skull size in the ‘‘Irish Elk’’ Megaloceros giganteus.
Alatalo RV, Hoglund J, Lundberg A, 1988. Patterns of variation in tail Evolution 28:191–220.
ornament size in birds. Biol J Linn Soc 34:363–374. Green AJ, 1992. Positive allometry is likely with mate choice,
Andersson M, 1994. Sexual selection. Princeton, New Jersey: Prince- competitive display and other functions. Anim Behav 43:170–172.
ton University Press. Gwynne DT, Jamieson I, 1998. Sexual selection and sexual dimorphism
Baker RH, Wilkinson GS, 2001. Phylogenetic analysis of sexual in a harem-polygynous insect, the alpine weta (Hemideina maori,
dimorphism and eye-span allometry in stalk-eyed flies (Diopsidae). Orthoptera Stenopelmatidae). Ethol Ecol Evol 10:393–402.
Evolution 55:1373–1385. Hanley RS, 2001. Mandibular allometry and male dimorphism in
Barrett P, 1991. Keeping wetas in captivity. Wellington: Wellington a group of obligately mycophagous beetles (Insecta: Coleoptera: