BioOne sees sustainable scholarly publishing as an inherently collaborative enterprise connecting authors, nonprofit publishers, academic institutions, researchlibraries, and research funders in the common goal of maximizing access to critical research.
Ecological determinants of clinal morphological variation in the cranium of theNorth American gray wolfAuthor(s): F. Robin O'Keefe , Julie Meachen , Elizabeth V. Fet , and Alexandria BrannickSource: Journal of Mammalogy, 94(6):1223-1236. 2013.Published By: American Society of MammalogistsDOI: http://dx.doi.org/10.1644/13-MAMM-A-069URL: http://www.bioone.org/doi/full/10.1644/13-MAMM-A-069
BioOne (www.bioone.org) is a nonprofit, online aggregation of core research in the biological, ecological, andenvironmental sciences. BioOne provides a sustainable online platform for over 170 journals and books publishedby nonprofit societies, associations, museums, institutions, and presses.
Your use of this PDF, the BioOne Web site, and all posted and associated content indicates your acceptance ofBioOne’s Terms of Use, available at www.bioone.org/page/terms_of_use.
Usage of BioOne content is strictly limited to personal, educational, and non-commercial use. Commercial inquiriesor rights and permissions requests should be directed to the individual publisher as copyright holder.
Journal of Mammalogy, 94(6):1223–1236, 2013
Ecological determinants of clinal morphological variation in thecranium of the North American gray wolf
F. ROBIN O’KEEFE,* JULIE MEACHEN, ELIZABETH V. FET, AND ALEXANDRIA BRANNICK
Department of Biological Sciences, Marshall University, One John Marshall Drive, Huntington, WV 25755, USA (FRO,EVF, AB)Joan C. Edwards School of Medicine, Marshall University, 1600 Medical Center Drive, Huntington, WV 25755, USA(JM)
* Correspondent: [email protected]
The gray wolf (Canis lupus) exhibits both genetic and morphologic clinal variation across North America.
Although shape variation in wolf populations has been documented, no study has been made to exhaustively
quantify it, or to correlate morphologic variation with environmental variables. This study utilizes a large
historical database of wolf skull linear measurements to analyze shape, and attempts to correlate it with wolf
ecology. A variety of statistical tests are employed; size and shape are examined through a principal component
analysis and a calculation of allometry vectors. Multiple regression analysis (both global and stepwise) are then
used to test the resulting principal components against various biotic and abiotic factors. In addition, the effects
of sexual dimorphism and taxonomy on morphology are explored through 1-way analysis of variance and
canonical variates analysis, respectively. Several patterns are revealed, including size increase with latitude in
accord with Bergmann’s rule. Static allometry is significant, the fundamental pattern being a decrease in the
robusticity of the basicranium relative to the viscerocranium. Sexual dimorphism, allometry, and a correlation
with precipitation are other key factors driving morphological variation. Examination of these patterns has
allowed us to make conclusions about the direct and indirect ways the environment has affected clinal variation
in wolves.
Key words: allometry, biogeography, carnivory, morphometrics
� 2013 American Society of Mammalogists
DOI: 10.1644/13-MAMM-A-069
The gray wolf (Canis lupus) may be the most familiar of
large mammalian carnivores. Its relationship with humans goes
back millennia and includes the origin of the domestic dog
(Vila et al. 1997; Wang and Tedford 2008), as well as a long
history of uneasy cohabitation, persecution, and coevolution
with humans (Schleidt and Shalter 2003). The gray wolf once
possessed the widest geographic range of any land mammal,
consisting of the entire Northern Hemisphere north of 13–208N
latitude (Boitani 2003), and encompassing the vast range of
environments found therein. This historical range has varied
widely over the last several thousand years due to changes in
human persecution patterns (Okarma 1993); currently, the
range of C. lupus is restricted to a far northern distribution,
having been extirpated in southern Europe, and in most of
North America south of the Canadian border (Leonard et al.
2005). However, active recolonization efforts are again
expanding this range (Paquet and Carbyn 2003).
Concerted efforts to destroy the gray wolf in North America
over the last 150 years have produced 1 unintended benefit.
Large museum collections of gray wolf pelts and skeletal
material were amassed during this time, and this material is
now available for study. Goldman (1944) utilized the products
of ‘‘predatory animal control work’’ undertaken by the United
States Biological Survey in his analysis and classification of
North American wolves. This sample numbered 1,368
individuals. The pioneering biometrician Pierre Jolicoeur
analyzed a more recently collected sample of 499 wolf skulls
from northwestern Canada in one of the 1st applications of
canonical variates analysis (CVA) to vertebrate biology
(Jolicoeur 1959, 1975). Lastly, the work of Nowak (1979) on
subspecific taxonomy of C. lupus relied originally on a sample
of 379 wolves, whereas a later analysis (Nowak 1995) utilized
580 skulls. Bogan and Mehlhop (1983) analyzed 160 crania
from the southwestern United States in their discussion of
w w w . m a m m a l o g y . o r g
1223
subspecies taxonomy in this region. In summary, a wealth of
material is available for the study of geographic variation in the
gray wolf.
The studies listed above are concerned primarily or solely
with skulls, and all contain data in a ‘‘traditional’’ morpho-
metric format, comprising linear measurements of various
features rather than landmark coordinates (Marcus 1990). The
primary aim of most has been taxonomic, attempting to
identify combinations of skull characters sufficient to diagnose
various subspecies. C. lupus is a geographically variable
species and conservation efforts have put a premium on
subspecies taxonomy (Nowak 1979, 1995, 2003). This
taxonomy has changed markedly over time. Goldman (1944)
identified 23 subspecies of C. lupus in North America, whereas
Hall (1981) accepted 24 subspecies as valid. However, as early
as 1959 Jolicoeur believed that wolf populations lacked
discrete morphological division, and this impression was
confirmed by later authors (reviewed by Nowak [2003]).
Studies analyzing shape variability in C. lupus in a non-
taxonomic context are rare (Jolicoeur 1959, 1975; Skeel and
Carbyn 1977). In these studies variation was found to be clinal,
body size was found to increase with latitude in accord with
Bergmann’s rule, and there were few distinct size differences
among populations. Sexual dimorphism was significant and
largely size related.
More recently, a flood of genetic evidence has added to the
discussion on wolf taxonomy, significantly clarifying the
biological architecture of the Canis species complex in North
America (Rutledge et al. 2012) and furnishing a definitive
taxonomy (Koblmuller et al. 2009; vonHoldt et al. 2011;
Chambers et al. 2012). These genetic analyses have shown that
gene flow among populations is uniformly high, and that
geographic distance is the best predictor of genetic distance.
Variation is broadly clinal with hybrid zones of varying width,
and the use of typological subspecies is probably misleading
(Wayne and Vila 2003). In general, North American C. lupus is
a cosmopolitan and mobile species with ample gene flow, yet
with significant, fitness-related clinal variation in morphology
(Smith et al. 1997; Munoz-Fuentes et al. 2009).
The source and maintenance of clinal variation has been the
subject of several recent studies. Genetic differentiation in
European (Pilot et al. 2006, 2012) and North American
(Carmichael et al. 2007; Munoz-Fuentes et al. 2009) wolves
appears to be driven by ecological factors, probably through
the mechanism of natal habitat-biased dispersal (Haughland
and Larsen 2004a, 2004b). Munoz-Fuentes et al. (2009) further
demonstrated that genetic differentiation is correlated with
morphological differences. Wolves from coastal British
Columbia, Canada, differ genetically and morphologically
from those from the continental interior, even though dispersal
distances are relatively short. A recent study examined the
influence of geography and genetics on wolf hunting behaviors
in Europe and found a correlation between clinal variation,
genetic signal, and prey preference (Pilot et al. 2012). Also,
landscape variation has been shown to affect prey-hunting
strategies in wolves (McPhee et al. 2012). Fitness-related
morphology also is labile chronologically; for instance,
Leonard et al. (2007) document the presence and extinction
of a specialized, robust gray wolf ecomorph in Siberia. All
these studies implicitly or explicitly implicate environmental
factors as determinants of shape, and all demonstrate that
wolves of the genus Canis are a widespread yet locally adapted
species.
However, no study to date has quantified clinal variation in
wolf skull shape across the whole of North America, nor has
the purported link between shape variation and climate been
formally tested. Here we perform such a study. We 1st quantify
morphological variation in wolf skulls across a broad
geographic area, then explore how this variation correlates
with environmental variables such as climate and precipitation,
as well as biological variables such as taxonomy and sex. We
began by compiling a cranial morphometric data set from a
large sample of North American gray wolves. These data were
culled from a historical study and comprise linear measure-
ments, and the sampled skulls span the original geographic
range from Mexico to Alaska. The data set includes locality
data (in latitude and longitude) for each skull. We then
appended modern, high-resolution climate data to the data set
TABLE 1.—Measurement descriptions quoted from Goldman (1944:409); see Fig. 1.
Measurement Description
Greatest length Length from anterior tip of premaxillae to the posterior point of inion in median line over
foramen magnum
Condylobasal length Length from anterior tip of premaxillae to posterior plane of occipital condyles
Zygomatic breadth Greatest distance across zygomata
Squamosal constriction Distance across squamosals at constriction behind zygomata
Width of rostrum Width of rostrum at constriction behind canines
Interorbital breadth Least distance between orbits
Postorbital constriction Least width of frontals at constriction behind postorbital process
Length of mandible Distance from anterior end of mandible to plane of posterior ends of angles, the right and
left sides measured together
Height of coronoid process Vertical height from lower border of angle
Maxillary toothrow, crown length Greatest distance from curved front of canine to back of cingulum of posterior upper molar
Upper carnassial, crown length (outer side), and crown width Anteroposterior diameter of crown on outer side, and transverse diameter at widest point
anteriorly
First upper molar, anteroposterior diameter, and transverse diameter Greatest anteroposterior diameter of crown on outer side, and greatest transverse diameter
Lower carnassial (crown length) Anteroposterior diameter of cingulum
1224 Vol. 94, No. 6JOURNAL OF MAMMALOGY
by referencing each location. This data set proves to be a rich
source of insight into the intrinsic and extrinsic factors driving
local wolf adaptation.
MATERIALS AND METHODS
The statistical strategy adopted here is to regress traditional
morphometric data against presumptive biotic and abiotic
forcing factors. Size and shape are investigated using standard
principal component analysis (PCA), along with multivariate
allometry. Because many of the abiotic factors of interest are
correlated, for example, latitude and temperature, we use
multiple regression modeling in an attempt to untangle the
influence of each factor on each morphological variable.
Stepwise regression is then used to construct a best-fit model
for each principal component (PC). Lastly, although modern
wolf taxonomy is no longer based solely on morphological
data, it is still pertinent to that question (Chambers et al. 2012).
We therefore performed a CVA on both the raw data and size-
adjusted data. Chambers et al. (2012) posit 4 valid subspecies
of C. lupus in North America: C. l. arctos, C. l. baileyi, C. l.nubilus, and C. l. occidentalis. They have raised the previous
subspecies C. l. lycaon to species status, now designated C.
lycaon (Chambers et al. 2012). However, this taxonomy is
contentious. In this paper we accept the most broadly
recognized taxonomy, where C. l. lycaon remains a subspecies
(Koblmuller et al. 2009; vonHoldt et al. 2011). In most taxon-
based analyses in this paper, C. l. arctos is excluded because of
low sample size (n ¼ 2).
Data Compilation
Morphometric data.—Acquisition of the raw data began
with manual entry of morphometric data from Goldman
(1944). Chapter 13 of that study gives measurement details
(pp. 408–410). As stated by Goldman (1944), all wolves
measured were adults, as determined by eruption of the
complete adult dentition. However, because adult dentition is
in place by a somatic age of about 6–7 months, whereas full
body stature is not reached until 12–14 months and mass can
continue to increase (Kreeger 2003), we assumed that
significant ontogenetic variation still exists within the data.
Although Goldman (1944) also recorded data for body mass
and coat color, these data were significantly more restricted in
scope, and so were excluded here in the interest of sample size.
A total of 15 cranial measurements were recorded by Goldman
FIG. 1.—Measurements used in this study, as described by Goldman (1944). Measurements are as follows: A) greatest length of cranium; B)
condylobasal length; C) zygomatic breadth; D) squamosal constriction; E) width of rostrum; F) interorbital breadth; G) postorbital constriction; H)
length of mandible; I) height of coronoid process; J) maxillary toothrow, crown length; K) upper carnassial, crown length (outer side); L) upper
carnassial, crown width; M) 1st upper molar, anteroposterior diameter; N) 1st upper molar, transverse diameter; and O) lower carnassial crown
length.
December 2013 1225O’KEEFE ET AL.—WOLF CRANIAL VARIATION
(1944; Table 1; Fig. 1). The data set constructed for this paper
contains data from 312 wolves. Two hundred eighty-nine of
these had no missing measurements. One hundred thirty-six of
the wolves were female; 176 were male. Seventeen of the
wolves had no recoverable locality data, and 3 of these also
were missing morphometric data. All of the multivariate results
reported here are derived from analysis of the 289 wolves with
complete data; the univariate results utilize all available
measurements. See Supporting Information S1 (DOI: 10.
1644/13-MAMM-A-069.S1) for complete data.
FIG. 2.—Geographical locations of all captures for wolves in this study, shown on a Mercator projection of North America. Latitude and
longitude are calculated from locality information in Goldman (1944) and Google Earth (https://maps.google.com/). The abiotic variable mean
annual temperature is plotted from Hijmans et al. (2005).
TABLE 2.—Summary of eigenanalysis of the craniometric data set for Canis lupus, n¼ 263. Ordination was performed on the correlation matrix
and principal component (PC) coefficients are standardized so that the sum of squared coefficients for each vector is 1. Note that PC 1 is
dominated by the size factor, with coefficients that are universally high and positive; however, there is significant shape change as well on this
axis. PC 2 clearly portrays tooth size against the remainder of the skull and is a clear signal of the continued hypertrophy of the skull with respect
to the teeth after dental maturity but before full somatic maturity. Interpretation of PC 3 is more difficult, but seems to indicate an increase in the
size of the frontal area of the brain, and modest tooth size increase, relative to the rest of the skull. This axis is correlated with precipitation and
probably driven by ecological factors. PC 4 is again correlated with sex and taxon, and records residual size-correlated shape variation.
PC 1 PC 2 PC 3 PC 4 PC 5
Eigenvalue 10.164 1.452 0.750 0.577 0.410
Percent 67.761 9.680 5.003 3.849 2.735
Cumulative percent 67.761 77.442 82.445 86.293 89.028
Greatest length 0.295 0.104 �0.148 �0.191 �0.120
Condylobasal length 0.295 0.053 �0.143 �0.289 �0.153
Zygomatic breadth 0.270 0.230 �0.159 0.134 0.094
Squamosal constriction 0.265 0.157 �0.234 0.146 �0.344
Width of rostrum 0.262 0.100 �0.162 0.307 0.324
Interorbital breadth 0.230 0.314 0.361 0.096 0.575
Postorbital constriction 0.168 0.379 0.747 �0.100 �0.326
Length of mandible 0.294 0.137 �0.118 �0.239 �0.120
Height of coronoid process 0.256 0.230 �0.235 0.183 0.170
Maxillary toothrow, crown length 0.287 �0.014 �0.105 �0.333 �0.114
Upper carnassial, crown length (outer side) 0.266 �0.277 0.085 0.169 0.030
Upper carnassial, crown width 0.223 �0.274 0.159 0.618 �0.379
First upper molar, anteroposterior diameter 0.219 �0.435 0.156 �0.289 0.297
First upper molar, transverse diameter 0.246 �0.368 0.125 �0.160 0.024
Lower carnassial crown length 0.264 �0.329 0.104 0.074 0.054
1226 Vol. 94, No. 6JOURNAL OF MAMMALOGY
Locality data.—Locality data also were acquired from
Goldman (1944). This process was necessarily imprecise;
Goldman (1944) generally cites municipalities for his locality
of collection. However, working on the assumption that most
wolves registered at a given place came from the general area,
we used the municipalities as locations. These locations were
entered into Google Maps (https://maps.google.com/) and the
latitude and longitude of each location were recorded (Fig. 2).
The quality of climate data available from modern high-
resolution databases is high, on the scale of kilometers to tens
of kilometers (Hijmans et al. 2005), so the error in location of
collection is probably the more significant source of error in the
data set. The home-range size of wolves is highly variable, but
range radii of 50 miles (~80 kilometers) are known to occur,
and some wolves migrate with prey species (Mech and Boitani
2003); therefore, we believe our locality data are a reasonable
approximation of a given wolf’s area or origin.
Abiotic data.—Climatic data were taken from the database
compiled by Hijmans et al. (2005) and downloaded as surfaces
for all of North America in ArcGIS (Environmental Systems
Research Institute [ESRI] 2011). The latitude and longitude of
capture recordings were then queried to this database, and the
climate data for each point were recorded. These variables
included mean yearly temperature (Tavg), maximum and
minimum monthly mean temperature (Tmax C and Tmin C,
respectively), yearly mean precipitation (YMP), and yearly
precipitation variance (YPV).
TABLE 3.—Multiple regression results for principal component (PC) scores regressed against abiotic variables and sex of Canis lupus. Abiotic
variables include yearly mean precipitation (YMP), yearly precipitation variance (YPV), latitude (Lat), longitude (Long), maximum monthly mean
temperature (Tmax C), minimum monthly mean temperature (Tmin C), mean yearly temperature (Tavg), and sex. All regressions have n¼ 263,
the number of wolves with complete shape and abiotic data. For details on the global multiple regressions and stepwise regression modeling see
‘‘Materials and Methods.’’ AIC¼Akaike information criterion. Significance at the P , 0.05 level is indicated by 1 asterisk. Significance at the P, 0.01 level is indicated by 2 asterisks.
Shape variable
global regression
Modeled
variables t-ratio Prob. . jtjStepwise regression
model F-ratio Prob. . F
PC 1 YMP �1.64 0.1032 PC 1 5.528 0.0195
R2 ¼ 0.7576 YPV 1.36 0.1743 P ¼ 5
F-ratio ¼ 91.6109 Lat 3.06 0.0025** R2 ¼ 0.7567 24.197 0
P , 0.0001** Long �2.54 0.0116* AIC ¼ 244.8258 20.185 0
Tmax C �2.48 0.0139*
Tmin C �2.53 0.0121*
Tavg 2.8 0.0055**
Sex 18.9 0.0001** 355.368 0
Taxon 7.32 0.0001** 55.76 0
PC 2 YMP �0.32 0.7487 PC 2
R2 ¼ 0.0643 YPV 0.07 0.9447 P ¼ 2
F-ratio ¼ 1.9231 Lat 2.2 0.0288* R2 ¼ 0.0404
P , 0.0492* Long �1.08 0.2797 AIC ¼ 94.2512 9.951 0.0018
Tmax C �0.81 0.4184
Tmin C �0.23 0.815
Tavg 1.53 0.1263
Sex 0.73 0.4684
Taxon �2.16 0.0319* 6.361 0.0123
PC 3 YMP �3.46 0.0006** PC 3 11.633 0.0008
R2 ¼ 0.1380 YPV 4.54 0.0001** P ¼ 4 22.722 0
F-ratio ¼ 4.4387 Lat 0.23 0.8148 R2 ¼ 0.1300
P , 0.0001** Long 3.02 0.0028** AIC ¼ �93.2149 13.773 0.0003
Tmax C �1.48 0.139 13.614 0.0003
Tmin C 0.74 0.46
Tavg �0.03 0.9726
Sex �0.96 0.3373
Taxon 0.15 0.8845
PC 4 YMP �1.98 0.0487* PC 4 6.349 0.0124
R2 ¼ 0.2872 YPV 3.2 0.0015** P ¼ 6 13.058 0.0004
F-ratio ¼ 11.2793 Lat �3.69 0.0003** R2 ¼ 0.2853 23.528 0
P , 0.0001** Long �0.38 0.7051 AIC ¼ �204.089
Tmax C 0.48 0.6299
Tmin C �0.40 0.6907 2.074 0.1511
Tavg �0.72 0.4737
Sex 3.6 0.0004** 13.248 0.0003
Taxon 2.52 0.0122* 7.25 0.0076
December 2013 1227O’KEEFE ET AL.—WOLF CRANIAL VARIATION
Data Analysis
Principal component analysis (PCA).—Given the number of
variables and their uniformly high covariance, eigenanalysis is
appropriate for reducing the dimensionality of the data,
allowing them to be summarized by a relatively small
number of composite variables. Because the variances of the
variables were heterogeneous, the correlation rather than
covariance matrix was used in the PCA (Reyment and
Joreskog 1996), except in the special case of calculation of
the allometry vector (see below). The results of the PCA are
reported in Table 2.
The PCA was performed on the raw data, without log
transformation. As pointed out by Jungers et al. (1995), the log
transform is overused in biological analysis and is often not
necessary, even when size variation is large. In this study,
exploratory analyses showed that ordinations (both PCA and
CVA) and multiple regressions run on log-transformed data
were essentially identical to those from the raw data. This
demonstrates that nonlinearities are not biasing the ordinations,
and that the log transform is not necessary.
Multiple regression.—Once the correlation structure of the
core morphometric matrix was determined, individual scores
for each PC deemed significant (PCs 1–4) were calculated and
appended to the data set. These PC scores are Mossiman shape
variables and are therefore permissible for use in regression
and other parametric statistical analysis techniques (Mosimann
and Malley 1979; Reyment 1991).
We explored the correlation of each PC with a range of
biotic and abiotic factors through the use of multiple
regression. The variables utilized in multiple regression were
yearly mean precipitation (YMP), yearly precipitation variance
(YPV), latitude (Lat), longitude (Long), maximum monthly
mean temperature (Tmax C), minimum monthly mean
temperature (Tmin C), mean yearly temperature (Tavg), sex
(male or female), and taxon (functionally this was 4 classes, the
valid subspecies of C. lupus save for C. l. arctos, whose n¼ 2).
These variables were entered into a multiple regression model,
and regressed on each of the 4 PCs of interest. Global
significance for each regression, and for significances for each
predictor variable, are reported in Table 3. We also
experimented with stepwise regression; the results of stepwise
modeling for PCs 1–4 were very similar to the global multiple
regression models. The best-fit stepwise regression model is
also reported in Table 3; the number of parameters participat-
ing in each model was determined by iterating the entry of
variables into the model and selecting the model with the
lowest Akaike information criterion (AIC).
Allometry and sexual dimorphism.—Given the amount of
size variation in the data set, and the documented existence of
sexual dimorphism within C. lupus (Goldman [1944] and more
recent references), we performed formal analyses to investigate
allometry and sexual dimorphism. To analyze allometry, we 1st
calculated the global (all taxa combined) multivariate allometry
vector introduced by Jolicoeur (1963). This vector is defined as
the 1st PC derived from eigenanalysis of the covariance matrix
of log-transformed variables (O’Keefe et al. 1999). The global
allometry vector, and its comparison with the vector of
isometry, is depicted in Fig. 3. Significance values for the
deviation of each coefficient from isometry were not calculated
by bootstrapping the eigenanalysis; significance was calculated
from regression of each univariate variable against the
geometric mean (GM) size estimator. Exploratory analysis
using other size estimators (PC 1 score, greatest length)
demonstrates that the results obtained are robust. In these
regressions and those discussed below, ordinary least squares
regression was used; however, because the R2 values were very
high, calculation of the reduced major axis slope was not
necessary.
Not surprisingly, the multivariate allometry vector differs
significantly from isometry (see ‘‘Results’’); however, we note
that because of the presence of significant late-stage somatic
growth (after the eruption of adult dentition—Kreeger 2003),
and the presence of different taxa, the global allometry vector
will contain phylogenetic, ontogenetic, and static components
(Klingenberg 1996). The presence of significant phylogenetic
allometry (that among taxa) is a possibility, and we tested for
this by regressing the allometry vector scores of each taxon
against the GM. This analysis is equivalent to a standard
FIG. 3.—Histogram depicting the multivariate allometry vector (the
1st eigenvector derived from the covariance matrix of ln-transformed
variables; see ‘‘Materials and Methods’’). Face, rostrum, and mandible
sizes increase with positive allometry, whereas tooth and brain sizes
increase with negative allometry. The coefficient of isometry for 15
variables is 0.258. Significance at the P , 0.01 level is indicated by 2
asterisks. Significance was determined by standard parametric reduced
major axis regression of bivariate allometry of each variable versus the
geometric mean.
1228 Vol. 94, No. 6JOURNAL OF MAMMALOGY
bivariate allometry regression of a given variable against size,
but the dependent variable in this case is the synthetic
multivariate allometry variable rather than a univariate
measurement. Because the GM is an isometric size estimator,
a significant difference from a slope of 1 in this regression
indicates that the growth trajectory of the sample varies
significantly from isometry. The actual regression in this case,
and in those described below, was performed with lnGM on the
abscissa and ln(AV þ 10), where AV is the allometry vector,
on the ordinate. The addition of a scalar to the allometry vector
scores is necessary because the ordination used to calculate the
scores is mean-centered to 0; therefore, half of the scores will
be less than 1, and their logarithms will be imaginary. All
details of allometry vector comparisons are reported in Table 4.
The global allometry vector also is correlated with sexual
dimorphism, because most (but not all) variation due to sex is
due to size and size-correlated shape variation. Sexual
dimorphism is an important part of the covariance structure
of the morphometric data, and is of obvious biological
significance. We therefore investigated the dimorphism in
each variable individually with simple 1-way analyses of
variance (ANOVAs) with sex as the independent variable;
results are presented in Table 5. An isometric size correction
was made by dividing each data row by its GM before the
ANOVA was performed.
Canonical variates analysis.—We employed the commonly
accepted taxonomy of the C. lupus species complex as a factor
in this study, including C. l. baileyi, C. l. nubilus, C. l.occidentalis, and C. l. lycaon. To investigate the impact of the
size factor, the CVA was run both on the raw data and on the
data with each value row normalized to its GM. Canonical
details are reported in Table 6.
RESULTS
Covariance Structure
Analysis of the resulting ordination via regression on
geographic location reveals north–south and east–west trends
in morphological variation. Regressions against temperature
and precipitation quantify morphological variation on a
continental scale. Table 2 reports the coefficients for the first
5 PCs; however, the 1st component carries 68% of the variance
in the correlation matrix. This, and the uniformly high and
positive coefficients on PC 1, indicate a large amount of size
variation in the sample, resulting in a robust size factor that is
TABLE 4.—A) Allometry vector calculations. All vectors were calculated as the 1st principal component of the covariance matrix derived from
ln-transformed data. Allometry vectors were then regressed on the geometric mean to yield an overall measure of allometry. The global vector
reflects positive allometry, being significantly different than 1. The vectors for each taxon also are significantly different, save for comparisons
with Canis lupus baileyi; this taxon has a small sample size and relatively small amount of size variation, and the vector is probably not
dependable. B) Allometry vectors for each taxon; the global allometry vector is identical to the one shown in Fig. 3.
A)
Taxon R2 P n Slope SE Lower 2 SE Upper 2 SE
Size
rank
C. l. baileyi 0.9965 0.0001 18 1.069 0.0158 1.0374 1.1006 3
C. l. lycaon 0.9978 0.0001 29 1.09 0.0099 1.0702 1.1098 3
C. l. nubilus 0.9969 0.0001 181 1.062 0.0044 1.0532 1.0708 2
C. l. occidentalis 0.9972 0.0001 59 1.033 0.0072 1.0186 1.0474 1
All 0.9978 0 289 1.053 0.0029 1.0472 1.0588
B)
Allometry vector Global
C. l.baileyi
C. l.lycaon
C. l.nubilus
C. l.occidentalis
Percent variance explained 64.4058 45.9529 64.5880 51.6576 60.0212
lnGreatest length 0.2625 0.1623 0.2485 0.2350 0.2175
lnCondylobasal length 0.2517 0.2162 0.2393 0.2285 0.1892
lnZygomatic breadth 0.2528 0.2346 0.2607 0.2917 0.2410
lnSquamosal constriction 0.2168 0.2007 0.1686 0.2427 0.1634
lnWidth of rostrum 0.3211 0.3027 0.3090 0.3275 0.3122
lnInterorbital breadth 0.3140 0.0456 0.3755 0.3643 0.4429
lnPostorbital constriction 0.2368 �0.2317 0.3430 0.1955 0.3103
lnLength of mandible 0.2811 0.1755 0.2683 0.2563 0.2196
lnHeight of coronoid process 0.3191 0.2952 0.2726 0.3143 0.3495
lnMaxillary toothrow, crown length 0.2327 0.1824 0.2508 0.2095 0.1805
lnUpper carnassial, crown length (outer side) 0.2367 0.3006 0.2077 0.2378 0.2467
lnUpper carnassial, crown width 0.2499 0.4472 0.2946 0.2533 0.2417
lnFirst upper molar, anteroposterior diameter 0.1946 0.1972 0.1377 0.2034 0.2040
lnFirst upper molar, transverse diameter 0.2282 0.3482 0.2041 0.2118 0.1629
lnLower carnassial crown length 0.2358 0.2873 0.1753 0.2346 0.2276
December 2013 1229O’KEEFE ET AL.—WOLF CRANIAL VARIATION
the strongest single driver behind the ordination. The 2nd PC
accounts for just under 10% of the variation, whereas the 3rd
accounts for 5%, PC 4 for less than 4%, and PC 5 for less than
3%. In a matrix of 15 completely uncorrelated variables, each
variable would account for about 7% of the variance in the
correlation matrix. Therefore only PCs 1 and 2 are greater than
the ‘‘strength’’ of 1 variable, whereas PC 3 is probably also of
interest, with PC 4 marginally so. However, in covariance
matrices with a very strong underlying factor driving
covariance (such as size), the eigenvalue of the 1st PC will
be large, and all others relatively small. In these cases, using
the criterion of the ‘‘strength’’ of 1 variable to determine the
significance of PCs may be misleading. (This criterion is
arbitrary in any case [Reyment 1991]; the true limiting
consideration in the interpretation of successively smaller
PCs is the magnitude of the measurement error in the original
data, a quantity that we cannot determine given the historical
nature of the data.) We further explored the wolf data by
running a PCA after dividing each row by its GM. This is an
isometric, approximate size correction, and removes much of
the size factor from the resulting ordination. The eigenvalues
from this PCA run 27.2%, 19.3%, 10.6%, 7.8%, and 5.56% for
PCs 1–5, respectively. In this size-corrected analysis, PCs 1–4
are all greater than the ‘‘strength’’ of 1 variable, whereas PC 5
and lower are not. Based on this consideration we offer
interpretations of the first 4 PCs below, although we note that
PCs 3 and 4 carry a relatively small proportion of the variance.
Principal component 1 size axis.—The results of the PCA
yielded 4 axes of possible interest. PC 1 (67% of variance
explained) is dominated by the size factor (Table 2) given that
all coefficients are large and positive. However, this vector is
not isometric, with significant size-related shape variation. In
general this shape variation is a less extreme version of the
variation carried on the allometry vector; however, much of the
ontogenetic negative allometry captured by the allometry
vector is captured by PC 2 in this ordination (see below). PC 1
is therefore a measure of general size, and also carries static
and phylogenetic allometry in the form of increased snout
length and decreased relative brain size as scores increase. PC
1 is highly correlated with GM and accounts for 67% of the
variance in the data set, indicating that the size factor is the
dominant contributor to the correlation structure. We would
anticipate PC 1 to correlate with any factor that might affect
body size.
Principal component 2 ontogeny axis.—Principal
component 2 accounts for 9.7% of the variance in the wolf
data set. For this axis, all measurements taken directly from the
teeth load against all other measurements. This axis therefore
contrasts the size of the teeth relative to the bony parts of the
skull. This signal is expected, because canids reach dental
maturity at about 6 or 7 months, whereas full somatic growth
can take a year or more (Kreeger 2003). An axis that contrasts
modest growth of the skull around static teeth should therefore
occur in all populations, and this signal is conserved among all
TABLE 5.—Summary of t-test on means of 1-way ANOVA with sex
as the independent variable for the variables shown. Significance at P, 0.01 is denoted with two asterisks. GM ¼ geometric mean; F ¼female, M ¼ male.
Variable Sex ANOVA
Greatest length/GM 0.593
Condylobasal length/GM 0.046* F . M
Zygomatic breadth/GM 0.004** M . F
Squamosal constriction/GM 0.346
Rostrum width/GM 0.0002** M . F
Interorbital breadth/GM 0.0009** M . F
Postorbital constriction/GM 0.008** F . M
Mandible length/GM 0.887
Coronoid height/GM 0.0001** M . F
Maxillary toothrow crown length/GM 0.0001** F . M
P4 crown length/GM 0.430
P4 crown width/GM 0.983
M1 anteroposterior diameter/GM 0.012* F . M
MI transverse diameter/GM 0.004** F . M
m1 crown length/GM 0.524
TABLE 6.—Results of canonical variates analysis (CVA) of
taxonomic units of Canis in North America. C. lupus arctos is not
included because of small sample size in the data set. Two analyses
were done, the 1st on the raw data (CVAR), the 2nd on size-
standardized data (CVAS). The results of both analyses are highly
significant. Part A of the table shows canonical details for both
analyses, and analytical details are: Raw canonical: Wilks’ k P-value ,
0.0001**, approximate F¼12.6269; size-adjusted canonical: Wilks’ k
P-value , 0.0001**, approximate F¼ 9.1110. C. l. baileyi differs the
most from all other units based on the measurements in this study,
with perfect classification in both analyses. C. l. lycaon is classified
successfully in 76% of cases with size included, and in 69% with size
removed. Classifications are shown in part B of the table. C. l. nubilusand C. l. occidentalis are relatively similar and segregate rather poorly.
Raw canonical: Wilks’ k P-value , 0.0001**, approximate F ¼12.6269; size-adjusted canonical: Wilks’ k P-value , 0.0001**,
approximate F ¼ 9.1110.
A)
Eigenvalue
Percent variance
explained
Canonical
correlation
CVAR 1 1.4091 61.7180 0.7648
CVAR 2 0.5802 25.4131 0.6059
CVAR 3 0.2938 12.8689 0.4765
CVAS 1 0.7625 49.0786 0.6577
CVAS 2 0.4838 31.1437 0.5710
CVAS 3 0.3073 19.7778 0.4848
B)
Actual/class
C. l.
baileyi
C. l.
lycaon
C. l.
nubilus
C. l.
occidentalis
Raw canonical
C. l. baileyi 18 0 0 0
C. l. lycaon 1 22 3 3
C. l. nubilus 10 21 137 13
C. l. occidentalis 0 3 7 49
Size-adjusted canonical
C. l. baileyi 18 0 0 0
C. l. lycaon 1 21 3 4
C. l. nubilus 12 17 118 34
C. l. occidentalis 1 4 9 45
1230 Vol. 94, No. 6JOURNAL OF MAMMALOGY
subspecies. Significantly, if this interpretation is correct, it
predicts that PC 2 should not be significantly correlated with
any outside variable, because it is an intrinsic pattern within the
growth of all wolves. This lack of correlation is exactly as
predicted (see below).
Principal component 3 ecological axis.—Interpretation of
PC 3 (5%) is more difficult. We know that it is ecologically
significant (see below), so it carries a coherent signal. Scoring
positively on this axis are interorbital breadth and postorbital
constriction; the tooth variables are only slightly positive,
whereas the others are negative. This axis seems to indicate a
broad forehead and large frontal part of the cranial cavity,
along with modest tooth size increase, against the rest of the
skull. There are no obvious a priori predications about the
correlation of extrinsic factors given the structure of this vector.
Principal component 4 axis.—This axis accounts for 3.9%
of the variance in the data set and is therefore of marginal
importance. It also is difficult to interpret; it seems to capture
residual shape variation due to size, given its regression results
(below). The coefficients reflect this, with high positive loading
for rostrum width and interorbital breadth, and large negative
loading for squamosal constriction and postorbital constriction.
This axis is therefore an index of basicranial size relative to
viscerocranial size; however, maxillary toothrow and mandible
length both load with the basicranial measures, whereas the
only tooth measurement with a strong loading is the width of
P4, again loading with the basicranium. Skulls measuring high
on this axis will therefore have a relatively small brain with
relatively wide forehead and snout, but the snout also is
relatively short.
Multiple Regression
Principal component 1 size axis.—As predicted, this vector
is highly correlated with intrinsic and extrinsic factors
influencing size (Table 3). One advantage of the PC
ordination is that, to a large degree, it segregates ontogenetic
allometry to PC 2. Therefore, the nonisometric aspects of PC 1
will reflect static allometry, phylogenetic allometry, and sexual
dimorphism. The global multiple regression is highly
significant; the 4 factors with the highest correlations are sex,
taxon, latitude, and Tavg, all of which have highly significantFIG. 4.—A) Principal component (PC) 1 calculated from the 15 raw
measurements. Complete data (shape plus locality) is available for 269
wolves; 284 wolves had complete shape data. Analysis was performed
on the correlation matrix; for eigenanalysis details see Table 2. All
coefficients on PC 1 are high and positive, indicating that PC 1 has a
large size component. The abscissa is latitude, a variable highly
correlated with PC 1 (see multivariate regression, Table 3). Body size
increases significantly with latitude, in accordance with Bergmann’s
rule. Sexual dimorphism also increases with latitude, as shown by the
significantly different slopes of male only (blue) and female only (red)
linear regression. Wolves are split by sex in the inset graph; in the
main graph they are split by taxonomy following Chambers et al.
(2012). Confidence intervals on slopes are P , 0.05, and are male-
only, female-only, and combined as indicated. Regression details are
as follows: male-only: R2¼0.423, PC 1¼�5.60þ0.153 Lat, slope SE
�6 0.015, P , 0.0001**; female-only: R2 ¼ 0.334, PC 1 ¼�8.17 þ0.129 Lat, slope SE 6 0.017, P , 0.0001**; combined: R2¼ 0.261,
PC 1¼�6.95þ 0.147 Lat, slope SE 6 0.015, P , 0.0001** (where
Lat is latitude). B) Principal component 3 plotted against yearly mean
precipitation (YMP), with which it is correlated. PC 3 records shape
change in response to aridity, and also is highly correlated with annual
precipitation variance and with longitude. Individuals that are more
positive on PC 3 have relatively large anterior brain size and a modest
increase in relative tooth size. Confidence interval on slope is P ,
0.05. Regression details are as follows: R2¼ 0.036, PC 3¼�0.924þ0.241 YMP, P , 0.0020**. Significance at the P , 0.05 level is
indicated by 2 asterisks.
December 2013 1231O’KEEFE ET AL.—WOLF CRANIAL VARIATION
P-values. We illustrate PC 1 by its bivariate regression on
latitude in Fig. 4. The wolves are split with respect to sex, and
it is clear that male mean general size is always larger than
female mean size. Also, there is a clear increase in mean body
size with latitude. This pattern has been identified by most
previous studies on wolf morphometrics that sample a
sufficiently wide latitude range, including Goldman (1944),
using these data. Goldman (1944) noted sexual dimorphism
and intertaxon size variability, and we found these correlations
as well, particularly the hypertrophy of the snout relative to the
basicranium. There was no correlation of PC 1 with
precipitation factors. The stepwise regression analysis returns
a strongly supported best-fit model with 5 parameters; sexual
dimorphism is by far the most important component of this
model, followed by taxonomic membership, and then
geographic location.
Principal component 2 ontogeny axis.—As predicted, the
axis representing ontogenetic increase in skull size relative to
the teeth shows almost no significant correlation with any
factor. The global regression is marginally significant at the
0.0492 level, and the R2 on the regression is weak (0.064). The
2 factors driving this weak correlation are latitude and
taxonomy. The stepwise regression model also is very weak,
giving a 2-parameter model comprising longitude and
taxonomy.
Principal component 3 ecological axis.—The multiple
regression for PC 3 is strong (P , 0.001, R2 ¼ 0.138). The
factors driving this correlation are yearly mean precipitation,
yearly precipitation, and longitude. Longitude is correlated
with the precipitation variables, and PC 3 is documenting
variation in response to differences in aridity. To demonstrate
this we plotted PC 3 against yearly mean precipitation in Fig. 4.
Significantly, the variation captured by PC 3 is not correlated
with taxonomy, and this trend is one that is common to all taxa
whose ranges encompass adequate variation in aridity. The
stepwise regression model is in complete accord, returning a
well-supported 4-parameter model comprising precipitation
mean, precipitation variance, longitude, and maximum
temperature.
Principal component 4 axis.—Principal component 4 has a
highly significant multiple regression (P , 0.001, R2¼ 0.287).
The variables driving this regression are sex, latitude, and the
precipitation variables, with a marginal contribution from
taxonomy. The stepwise regression analysis returns a well-
supported model of 6 parameters, with contributions from sex and
taxonomy, but also both precipitation variables as well as latitude
and minimum temperature. This axis appears to be a hybrid,
recording residual size-related variation but also carrying some
ecological signal. It is therefore difficult to interpret because the
underlying factors are confounded.
Allometry
Allometry was 1st analyzed as a pooled vector, and then
split out by taxon for comparison of allometry vectors (Table
4). Although there is significant variance in direction of the
allometry vector among taxa, the overall pattern in all taxa is
similar, and is well summarized by the pooled vector (Fig. 3).
The multivariate allometry vector is positive in aggregate,
differing significantly from isometry; however, there is a mix
of positively and negatively allometric variables obscured by
this general trend. Variables changing with significant negative
allometry include squamosal constriction, maxillary crown
length, and both measures of M1. The other tooth variables
also are negatively allometric, but not significantly so.
Positively allometric variables include rostrum width, interor-
bital breadth, mandible length, and coronoid process height.
Together these variables document an increase in viscerocra-
nium size at the expense of the basicranium and teeth; negative
allometry of brain size relative to the rest of the skull is a well
known feature of tetrapod ontogeny in general (Goodrich
1930). The negative allometry in the teeth is similar to that
reflected in PC 2, and is a consequence of the rest of the skull
‘‘growing up’’ around the teeth. Lastly, negative allometry in
M1 and positive allometry in coronoid height reflects sexual
dimorphism. General allometry in wolf skulls can therefore be
characterized as an increase in the size of the viscerocranium
relative to the brain and dentition.
Further analysis demonstrates significant differences in the
allometry vectors among taxa (Table 4). These differences are
relatively minor except for the vector of C. l. baileyi. This
vector has 1 negative eigenvalue, probably resulting from the
impact of an outlier on the small sample size (n¼ 18). Further
inspection of the data for C. l. baileyi shows that there is
relatively less size variance in this sample when compared to
the other 3 taxa. The standard deviation on its slope also is very
wide, making statistical comparisons fruitless. We therefore do
not consider C. l. baileyi further.
Sexual Dimorphism
Most of the sexual dimorphism among wolves is linked
directly to body size; the overall trend of relative increase in the
viscerocranium versus the basicranium (PC 1) is correlated
with sex more highly than with any other factor (Table 3).
Goldman recognized this in his 1944 study. However, there are
more subtle differences between males and females in the data
set, as revealed by the univariate ANOVAs (Table 5) and the
allometry vectors. Measures of the skull that vary between the
sexes clearly reflect the relative size of the viscerocranium
versus the basicranium. However, the coronoid process is
significantly taller in males, whereas the maxillary toothrow is
longer and M1 is larger in females. The magnitude of sexual
dimorphism also varies with latitude (Fig. 4); the slope of male
size increase with latitude is 0.1532 (6 0.015 SE, n ¼ 151),
whereas that of females is 0.1287 (6 0.017, n ¼ 118).
However, the error on the slopes is such that this trend is not
statistically significant.
Canonical Variates
Results of the CVA are highly significant, with the 15
measurements reported by Goldman (1944) forming a good
basis for taxonomic segregation in aggregate. Both the CVA on
1232 Vol. 94, No. 6JOURNAL OF MAMMALOGY
the raw data (CVAR) and that on size-adjusted data (CVAS)
give 3 significant discriminant axes. The CVAR 1 (for
ordination containing size) explains about 62% of the
taxonomic variation in the data set. This is similar to the
coefficient in PC 1, and as we know from the regression
analysis and from Goldman (1944), size is a major source of
variation among taxa. As expected, C. l. occidentalis, the taxon
with the largest body size, is segregated from the other taxa on
this axis, with the other taxa descending in size rank at lower
scores on this axis. Inspection of the bivariate plot shows that
the variables greatest length and mandible length are important
on this axis, demonstrating that the previously encountered
allometry is here as well. The 2nd discriminant function of this
ordination differentiates C. l. nubilus from C. l. baileyi and C. l.lycaon, with C. l. occidentalis in an intermediate position. The
bivariate plot for this axis indicates that rostrum width and the
M1 transverse diameter are important in the positive direction,
whereas P4 crown length and condylobasal length are
important in the negative direction. This implies that C. l.nubilus, and to a lesser extent C. l. occidentalis, has a wider
snout, a broader M1, and a smaller P4 relative to the smaller
taxa (C. l. lycaon and C. l. baileyi).The 2nd ordination, with isometric size removed via division
by the GM, is very similar to the 1st. Because the size
correction was isometric, allometric shape variance is still
found in this ordination, but because isometric size variation
has been removed the size factor no longer dominates the
ordination. The percent variance explained is therefore less for
the 1st discriminate function, and greater in the 2nd, when
compared to those of the raw CVA. However, a rigid rotation
of the ordination (Fig. 5) shows that the overall ordination and
segregation are quite similar.
DISCUSSION
The analysis of the covariation among measured variables of
the skull presented here reproduces several patterns that have
already been documented for wolves, such as sexual
dimorphism and change in size with latitude. However, we
present quantitative treatments of these phenomena across a
broad geographic area for the 1st time, allowing more precise
statements about these patterns, particularly in the multiple and
stepwise regressions with PC 1, analysis of the allometry
vectors, and examination of sexual dimorphism. We also
identify novel axes of variation that are of interest, particularly
an ontogenetic allometry axis (PC 2) and an axis that is linked
with precipitation (PC 3). Lastly, discriminant function
analysis yields a picture of significant but clinal variation,
with axes that are highly significant yet fail to achieve complete
segregation. The variables reported here are good segregators
of the 4 taxa considered. However, the taxa do not form
discreet clusters in discriminant space, and there are significant
FIG. 5.—Canonical variates analysis of raw (CVAR, plot A) and size-adjusted (CVAS, plot B) wolf data. Both discriminant functions are
highly significant even though size is a good segregator; for univariate taxonomic principal component analysis (PCA) ANOVAs see Table 4. The
15 variables reported here are fairly good estimators of species membership, although only Canis lupus baileyi is classified correctly in all cases (n¼ 18); see Table 5 for canonical details. Note that CVAR 1 is correlated with sex (Kruskal–Wallis): nonparametric rank sums test, P , 0.0001.
Sexual variation is relegated to correlation with only CVAS 3 in the size-adjusted analysis (not shown); Kruskal–Wallis P , 0.0001.
December 2013 1233O’KEEFE ET AL.—WOLF CRANIAL VARIATION
numbers of failures of discrimination for both the CVAR and
CVAS (Table 6). Failure to discriminate cannot be attributed to
lack of statistical power, because the samples here are
relatively large. Also, the taxon with the smallest sample size
(C. l. baileyi, n ¼ 18) is in fact the only taxon with perfect
discrimination. The pattern seen in the discrimination is
probably real, with significant size and shape overlap among
taxa. One would expect this in a species complex with broadly
clinal variation and significant zones of hybridization among
taxa, as is the case for the Canis complex (Chambers et al.
2012).
One serendipitous property of the PCA is that it yields a
clean ontogenetic allometry axis as PC 2. The signal of skull
growth relative to the teeth on this axis is unambiguous, and
the lack of significant correlation in the multiple regression
demonstrates that this variation is common to all wolves
independent of size, taxon, or sex. We can say more about PC
1, because we can assume that ontogeny is a minor component
of its variance. Therefore, allometric change in PC 1 is
attributable to static allometry (difference in adult size), sexual
dimorphism, and phylogenetic allometry. All of these factors
are highly correlated with PC 1, making this axis a good
representation of how shape changes with body size increase
across latitude, taxon, and sex. The pattern on PC 1 is clear,
with relative decrease in the basicranium and increasing
robusticity in the viscerocranium. This trend is common
among taxa and clinal with respect to taxon and sex; this is in
accord with Goldman’s (1944) original findings and those of
later authors, up to and including Chambers et al. (2012).
Unlike many other taxa (Geist 1987), the members of the
genus Canis reported here conform strongly to Bergmann’s
rule, with increase in size at higher latitudes and lower
temperatures (Blanckenhorn et al. 2006, and references
therein). Geist (1987) found that Bergmann’s rule in wolves
has less to do with temperature than with food availability;
north of 658N latitude, wolves show a decreasing size trend
correlated with smaller prey populations. Wolves also display
Rensch’s rule, with sexual dimorphism increasing with
latitude. However, given the amount of scatter within males
and females, this trend was not statistically significant. The
amount of scatter in Fig. 4 (PC 1 versus latitude) is itself
significant; wolves in general show much size variation and
overlap between large females and small males. Some of this
scatter may be due to a lack of precision in our locality
estimates; however, we doubt that this is large enough to
account for a significant portion of the variation seen. The
members of the genus Canis reported here are variable, and this
appears to be a trait of their population biology. The variation
is broadly clinal and strongly linked to sex, and is probably
maintained by extensive hybridization in hybrid zones between
taxa (Chambers et al. 2012).
MacNulty et al. (2009) also noted that male wolves continue
to increase in body size until age 4.75 years, thereafter showing
a decline in body size. However, female wolves did not show
this decline, but continue to increase in size throughout their
lives. Although male skull size would likely not decrease,
female wolves may continue to show minor but measureable
skeletal growth far into adulthood, as has been demonstrated in
some other carnivorans (Binder and Van Valkenburgh 2000;
Meachen-Samuels and Binder 2010). This may contribute to
the overlap in skull size between males and females.
Principal component 3 demonstrates that skull morphology
is correlated with precipitation. Wolves in the wettest areas
show larger frontal bones and larger (longer and wider) teeth.
This finding may be correlated with increased prey acquisition
in areas with higher primary productivity associated with high
precipitation. A recent study on Japanese deer found that body
size was positively correlated with precipitation, and that these
areas also had the highest productivity (Terada et al. 2012).
Similar results also were found in woodrats (Cordero and Epps
2012) and ground squirrels (Gur 2010). This indicates that
precipitation is an important factor impacting prey size, and
that its effects are visible in wolves. However, the exact
mechanics driving this correlation are unclear, and a topic of
further research.
We did identify several single measures that varied
significantly between the sexes (Table 5). Most of these are
attributable to static allometry as described above; however,
males have significantly taller coronoid processes, whereas
females have significantly larger M1s. However, Gittleman and
Van Valkenburgh (1997) found that the carnassial teeth (lower
m1 and upper P4; the primary meat-shearing teeth in
carnivorans) showed less sexual dimorphism than the canines
in many carnivorans, indicating that females and males use
their shearing teeth roughly equally for food processing. We
replicated this finding, but the difference in molar size implies
a sex-based difference in use of the more-posterior grinding
teeth.
Large molars and a relatively short coronoid process may
indicate more masseter use in females than in males, involving
more food processing and less prey acquisition. Subsidiary
analyses demonstrate that this is not a pure size pattern;
relatively small males still have tall coronoids and small molars
when compared to females of equivalent size. MacNulty et al.
(2009) found that prey acquisition and body size were
positively correlated; larger males generally outperform
females when handling prey. This may influence the feeding
behavior of wolves, causing females to consume prey quickly
to assure they can get their fill. Additionally, females may have
larger teeth overall to put them on par or ahead of males while
feeding if they fall short in pursuit and prey killing. Finally, we
speculate that relatively large female grinding dentition also
may be due to more complete carcass processing due to
increased nutrient needs during lactation. Sexual dimorphism is
thus a vector carrying size variation, with the basicranial–
viscerocranial trade-off implicit in all size change in Canis, but
also carrying positive allometry in coronoid height and
negative allometry in the size of M1.
Assuming that fossil wolf biology was similar to that of
extant wolves, these metrics also may be useful for establishing
the sex of fossil wolves. Given the amount of overlap between
the sexes, perfect discrimination is not possible, but samples of
1234 Vol. 94, No. 6JOURNAL OF MAMMALOGY
wolves with significant statistical power should show a
negative correlation between body size and molar size and a
positive correlation between coronoid height and body size.
Analysis of the multivariate allometry vectors is more
complex than that of the PC vectors. The reason for this is that
the coefficients for the allometry vector are similar to those of
PC 1, but also carry an overprint of ontogenetic allometry as
well. This signal resides in PC 2 in the PCA. Study of the
allometry vectors is hampered by the small sample size for C. l.baileyi. However, the allometry vectors of the other 3 taxa
show an interesting trend, with C. l. lycaon having the steepest
slope relative to isometry, followed by C. l. nubilus, then C. l.occidentalis. This also is the rank order of mean body size in
reverse; C. l. lycaon is the smallest in size, whereas C. l.occidentalis is the largest. This pattern might be explainable by
the populations growing toward a similar adult shape, but those
with smaller size having to adopt more extreme allometry to do
so. In terms of Gould’s clock model (Gould 1977) this would
be an example of ‘‘proportioned dwarfism.’’ If one envisions
what proportioned dwarfism would mean in vector space, it
entails greater shape change over a smaller range of body size.
This requires a vector with a more positively allometric slope,
whereas proportioned gigantism would require an isometric
vector. One also may predict that the coefficients of the
allometry vector would change in a regular way, with
allometric measures approaching isometry as body size
increases. However, this is not the pattern seen when
comparing the allometry vectors among taxa, so a simple
Gouldian interpretation of this heterochrony breaks down. In
fact the adult shapes of the different taxa do differ, and the
ontogenetic trajectories used to get to them vary in a complex
manner; this type of complex heterochrony has been observed
in reptiles as well (O’Keefe et al. 1999).
Using a large data set, we have shown that wolf skull
morphology is indirectly influenced by both temperature
(Bergmann’s rule) and precipitation, and directly influenced
by prey availability and primary productivity. Wolf subspecies
show clinal variation across subspecies and static and
phylogenetic allometry play a large role in the differences
among these groups. Our analyses showed that the differences
between the sexes have a large allometric component and we
were able to tease this out of our data using PCA and
multivariate allometry vectors. Static allometry seems to be the
main morphological component that is changing in wolves,
allowing for small concurrent changes in ecology, such as
differences in feeding ecology found in disparate populations
of grey wolves in Europe (Pilot et al. 2012). In addition,
landscape heterogeneity plays a role in how wolves choose to
hunt and kill prey as well (McPhee et al. 2012), which can
thereby shape morphological changes via adaptation, as we
observed in the clinal variation related to precipitation. Lastly,
sexual dimorphism is considerable, but males and females
overlap widely, and most of the intersex variation is confined
to the static allometry axis. However, sexual differences in
coronoid height and molar size hint at difference in hunting
success and carcass utilization between males and females.
ACKNOWLEDGMENTS
This study was supported by funds from the Marshall Foundation
and a grant from the National Aeronautics and Space Administration
West Virginia Space Grant Consortium to FRO. B. Van Valkenburgh
provided valuable input during the gestation of this paper.
SUPPORTING INFORMATION
SUPPORTING INFORMATION S1.—Data analyzed in this paper.
Found at DOI: 10.1644/13-MAMM-A-069.S1
LITERATURE CITED
BINDER, W. J., AND B. VAN VALKENBURGH. 2000. Development of bite
strength and feeding behaviour in juvenile spotted hyenas (Crocutacrocuta). Zoological Journal of the Linnean Society (London)
252:273–283.
BLANCKENHORN, W. U., R. C. STILLWELL, K. A. YOUNG, C. W. FOX, AND
K. G. ASHTON. 2006. When Rensch meets Bergmann: does sexual
size dimorphism change systematically with latitude? Evolution
60:2004–2011.
BOGAN, M. A., AND P. MEHLHOP. 1983. Systematic relationships of
gray wolves (Canis lupus) in southwestern North America.
Occasional Papers of the Museum of Southwestern Biology 1:1–21.
BOITANI, L. 2003. Wolf conservation and recovery. Pp. 317–340 in
Wolves: behavior, ecology, and conservation (L. D. Mech and L.
Boitani, eds.). University of Chicago Press, Chicago, Illinois.
CARMICHAEL, L. E., ET AL. 2007. Historical and ecological determinants
of genetic structure in arctic canids. Molecular Ecology 16:3466–
3483.
CHAMBERS, S. M., S. R. FAIN, B. FAZIO, AND M. AMARAL. 2012. An
account of the taxonomy of North American wolves from
morphological and genetic analyses. North American Fauna 77:1–
67.
CORDERO, G. A., AND C. W. EPPS. 2012. From desert to rainforest:
phenotypic variation in functionally important traits of bushy-tailed
woodrats (Neotoma cinerea) across two climatic extremes. Journal
of Mammalian Evolution 19:135–153.
ENVIRONMENTAL SYSTEMS RESEARCH INSTITUTE (ESRI). 2011. ArcGIS
Desktop: release 10. Environmental Systems Research Institute,
Redlands, California.
GEIST, V. 1987. Bergmann’s rule is invalid. Canadian Journal of
Zoology 65:1035–1038.
GITTLEMAN, J. L., AND B. VAN VALKENBURGH. 1997. Sexual
dimorphism in the canines and skulls of carnivores: effects of size,
phylogency, and behavioural ecology. Journal of Zoology 242:97–
117.
GOLDMAN, E. A. 1944. Classification of wolves. Pp. 389–636 in The
wolves of North America, part 2 (S. P. Young and E. A. Goldman,
eds.). Dover Publications, Inc., New York.
GOODRICH, E. S. 1930. Studies on the structure and development of
vertebrates. Macmillan and Co., London, United Kingdom.
GOULD, S. J. 1977. Ontogeny and phylogeny. Belknap Press of
Harvard University Press, Cambridge, Massachusetts.
GUR, H. 2010. Why do Anatolian ground squirrels exhibit a
Bergmannian size pattern? A phylogenetic comparative analysis
of geographic variation in body size. Biological Journal of the
Linnean Society 100:695–710.
HALL, E. R. 1981. The mammals of North America. John Wiley &
Sons, Inc., New York.
HAUGHLAND, D. L., AND K. W. LARSEN. 2004a. Ecology of NorthAmerican red squirrels across contrasting habitats: relating nataldispersal to habitat. Journal of Mammalogy 85:225–236.
December 2013 1235O’KEEFE ET AL.—WOLF CRANIAL VARIATION
HAUGHLAND, D. L., AND K. W. LARSEN. 2004b. Exploration correlateswith settlement: red squirrel dispersal in contrasting habitats.Journal of Animal Ecology 73:1024–1034.
HIJMANS, R. J., S. E. CAMERON, J. L. PARRA, P. G. JONES, AND A. JARVIS.2005. Very high resolution interpolated climate surfaces for globalland areas. International Journal of Climatology 25:1965–1978.
JOLICOEUR, P. 1959. Multivariate geographical variation in the wolfCanis lupus L. Evolution 13:283–299.
JOLICOEUR, P. 1963. The multivariate generalization of the allometryequation. Biometrics 19:497–499.
JOLICOEUR, P. 1975. Sexual dimorphism and geographical distance asfactors of skull variation in the wolf Canis lupus L. Pp. 54–61 inThe wild canids: their systematics, behavioral ecology andevolution (M. W. Fox, ed.). Van Nostrand Reinhold Co., NewYork.
JUNGERS, W. L., A. B. FALSETTI, AND C. E. WALL. 1995. Shape, relativesize, and size-adjustment in morphometrics. Yearbook of PhysicalAnthropology 38:137–161.
KLINGENBERG, P. C. 1996. Multivariate allometry. Pp. 23–49 inAdvances in morphometrics 284 (L. F. Marcus, M. Corti, A. Loy,G. J. Naylor, and D. E. Slice, eds.). Plenum Press, New York.
KOBLMULLER, S., M. NORD, R. K. WAYNE, AND J. A. LEONARD. 2009.Origin and status of the Great Lakes wolf. Molecular Ecology18:2313–2326.
KREEGER, T. J. 2003. The internal wolf: physiology, pathology, andpharmacology. Pp. 239–258 in Wolves: behavior, ecology, andconservation (L. D. Mech and L. Boitani, eds.). University ofChicago Press, Chicago, Illinois.
LEONARD, J. A., C. VILA, K. FOX-DOBBS, P. L. KOCH, R. WAYNE, AND
VAN VALKENBURGH. B. 2007. Megafaunal extinctions and thedisappearance of a specialized wolf ecomorph. Current Biology17:1146–1150.
LEONARD, J. A., C. VILA, AND R. K. WAYNE. 2005. Legacy lost: geneticvariability and population size of extirpated US grey wolves (Canislupus). Molecular Ecology 14:9–17.
MACNULTY, D. W., L. SMITH, D. MECH, AND L. E. EBERLY. 2009. Bodysize and predatory performance in wolves: is bigger better? Journalof Animal Ecology 78:532–539.
MARCUS, L. 1990. Traditional morphometrics. Pp. 77–122 inProceedings of the Michigan Morphometrics Workshop (F. J.Rohlf and F. L. Bookstein, eds.). Special Publication 2. Universityof Michigan Museum of Zoology, Ann Arbor.
MCPHEE, H. M., N. F. WEBB, AND E. H. MERRILL. 2012. Time-to-kill:measuring attack rates in a heterogeneous landscape with multipleprey types. Oikos 121:711–720.
MEACHEN-SAMUELS, J., AND W. J. BINDER. 2010. Sexual dimorphismand ontogenetic growth in the American lion and sabertoothed catfrom Rancho La Brea. Journal of Zoology (London) 280:271–279.
MECH, L. D., AND L. BOITANI. 2003. Wolf social ecology. Pp. 1–34 inWolves: behavior, ecology, and conservation (L. D. Mech and L.Boitani, eds.). University of Chicago Press, Chicago, Illinois.
MOSIMANN, J. E., AND J. D. MALLEY. 1979. Size and shape variables.Pp. 175–189 in Multivariate methods in ecological work (L. Orioci,C. R. Rao, and W. M. Stiteler, eds.). International CooperativePublishing House, Fairland, Maryland.
MUNOZ-FUENTES, V., C. T. DARIMONT, R. K. WAYNE, P. C. PAQUET, AND
J. A. LEONARD. 2009. Ecological factors drive differentiation inwolves from British Columbia. Journal of Biogeography 36:1516–1531.
NOWAK, R. M. 1979. North American Quaternary Canis. Monograph6. Museum of Natural History, University of Kansas, Lawrence.
NOWAK, R. M. 1995. Another look at wolf taxonomy. Pp. 375–397 inEcology and conservation of wolves in a changing world (L. N.Carbyn, S. H. Fritts, and D. R. Seip, eds.). Canadian CircumpolarInstitute, Edmonton, Alberta, Canada.
NOWAK, R. M. 2003. Wolf evolution and taxonomy. Pp. 239–258 inWolves: behavior, ecology, and conservation (L. D. Mech and L.Boitani, eds.). University of Chicago Press, Chicago, Illinois.
OKARMA, H. 1993. Status and management of the wolf in Poland.Biological Conservation 66:153–158.
O’KEEFE, F. R., O. RIEPPEL, AND P. M. SANDER. 1999. Shapedisassociation and inferred heterochrony in a clade of pachypleur-osaurs (Reptilia, Sauropterygia). Paleobiology 25:504–517.
PAQUET, P. C., AND L. N. CARBYN. 2003. Wolf, Canis lupus. Pp. 482–510 in Wild mammals of North America: biology, management,and conservation (G. A. Feldhamer, B. C. Thompson, and J. A.Chapman, eds.). Johns Hopkins University Press, Baltimore,Maryland.
PILOT, M., ET AL. 2006. Ecological factors influence population geneticstructure of European grey wolves. Molecular Ecology 15:4533–4553.
PILOT, M., W. JEDRZEJEWSKI, V. E. SIDOROVICH, W. MEIER-AUGENSTEIN,AND A. R. HOELZEL. 2012. Dietary differentiation and the evolutionof population genetic structure in a highly mobile carnivore. PLoSONE 7:e39341.
REYMENT, R. A. 1991. Multidimensional paleobiology. Pergamon,Oxford, United Kingdom.
REYMENT, R., AND K. G. JORESKOG. 1996. Applied factor analysis in thenatural sciences. Cambridge University Press, Cambridge, UnitedKingdom.
RUTLEDGE, L. Y., P. J. WILSON, C. F. KLUTSCH, B. R. PATTERSON, AND B.N. WHITE. 2012. Conservation genomics in perspective: a holisticapproach to understanding Canis evolution in North America.Biological Conservation 155:186–192.
SCHLEIDT, W. M., AND M. D. SHALTER. 2003. Co-evolution of humansand canids: an alternative view of dog domestication: homo hominilupus? Evolution and Cognition 9:57–72.
SKEEL, M. A., AND L. N. CARBYN. 1977. The morphologicalrelationship of gray wolves (Canis lupus) in national parks ofcentral Canada. Canadian Journal of Zoology 55:737–747.
SMITH, D., ET AL. 1997. Is incest common in gray wolf packs?Behavioral Ecology 8:384–391.
TERADA, C., S. TATSUZAWA, AND S. TAKASHI. 2012. Ecologicalcorrelates and determinants of deer morphology. Oecologia169:981–994.
VILA, C., ET AL. 1997. Multiple and ancient origins of the domesticdog. Science 276:1687–1689.
VONHOLDT, B. M., ET AL. 2011. A genome-wide perspective on theevolutionary history of enigmatic wolf-like canids. GenomeResearch 21:1294–1305.
WANG, X., AND R. H. TEDFORD. 2008. Dogs: their fossil relatives andevolutionary history. Columbia University Press, New York.
WAYNE, R. K., AND C. VILA. 2003. Molecular and genetic studies ofwolves. Pp. 218–238 in Wolves: behavior, ecology, and conserva-tion (L. D. Mech and L. Boitani, eds.). University of Chicago Press,Chicago, Illinois.
Submitted 14 March 2013. Accepted 22 July 2013.
Associate Editor was Ryan W. Norris.
1236 Vol. 94, No. 6JOURNAL OF MAMMALOGY