THE EVOLUTIONARY ROLE OF MODULARITY AND INTEGRATION … · THE EVOLUTIONARY ROLE OF MODULARITY AND INTEGRATION IN THE HOMINOID CRANIUM Philipp Mitteroecker1,2,3,4 and Fred Bookstein1,3,5

ORIGINAL ARTICLE

doi:10.1111/j.1558-5646.2008.00321.x

THE EVOLUTIONARY ROLE OF MODULARITYAND INTEGRATION IN THE HOMINOIDCRANIUMPhilipp Mitteroecker1,2,3,4 and Fred Bookstein1,3,5

1Department of Anthropology, University of Vienna, Althanstrasse 14, A-1091 Vienna, Austria2E-mail: [email protected]

3Konrad Lorenz Institute for Evolution and Cognition Research, Adolf Lorenz Gasse 2, A-3422 Altenberg, Austria4Department of Theoretical Biology, University of Vienna, Althanstrasse 14, A-1091 Vienna, Austria5Department of Statistics, University of Washington, Padelford B-207, Seattle, Washington 98195

Received June 27, 2007

Accepted December 19, 2007

Patterns of morphological integration and modularity among shape features emerge from genetic and developmental factors

with varying pleiotropic effects. Factors or processes affecting morphology only locally may respond to selection more easily than

common factors that may lead to deleterious side effects and hence are expected to be more conserved. We briefly review evidence

for such global factors in primate cranial development as well as for local factors constrained to either the face or the neurocranium.

In a sample comprising 157 crania of Homo sapiens, Pan troglodytes, and Gorilla gorilla, we statistically estimated common and

local factors of shape variation from Procrustes coordinates of 347 landmarks and semilandmarks. Common factors with pleiotropic

effects on both the face and the neurocranium account for a large amount of shape variation, but mainly by extension or truncation

of otherwise conserved developmental pathways. Local factors (modular shape characteristics) have more degrees of freedom for

evolutionary change than mere ontogenetic scaling. Cranial shape is similarly integrated during development in all three species, but

human evolution involves dissociation among several characteristics. The dissociation has probably been achieved by evolutionary

alterations and by the novel emergence of local factors affecting characteristics that are controlled at the same time by the common

factors.

KEY WORDS: Craniofacial growth, factor analysis, geometric morphometrics, heterochrony, human evolution, morphological

integration, partial least squares.

Modularity is a property of complex structures or processes thatare experimentally or conceptually separable into several “near-decomposable” (Simon 1962) modules. Following the lead ofNeedham (1933), Riedl (1978), and others, the concept of modu-larity has been applied in both developmental biology and evolu-tionary biology. Recent reviews can be found, for example, in Raff(1996), von Dassow and Munro (1999), Bolker (2001), Schlosserand Wagner (2004), and Callebaut and Rasskin-Gutman (2005). Inthe context of evolutionary biology, a module is usually construedas a set of morphological characters that (1) collectively serve

a common functional role, (2) are tightly integrated by strongpleiotropic effects of genetic variation, and (3) are relatively inde-pendent from other modules. Wagner (1996) and Wagner and Al-tenberg (1996) described such a modular organization in terms of a“genotype–phenotype map” characterizing how genetic variationcontributes to phenotypic variation throughout ontogeny (Fig. 1,left). A modular genotype–phenotype map can be decomposedinto several independent and more local genotype–phenotypemaps with fewer pleiotropic effects among the modules. If ge-netic changes affect only a part of the phenotype, the genome can

943C! 2008 The Author(s). Journal compilation C! 2008 The Society for the Study of Evolution.Evolution 62-4: 943–958

P. MITTEROECKER AND F. BOOKSTEIN

Figure 1. Left: A genotype–phenotype map visualizing the effectsof the genes G1 . . . G6 on the phenotypic characters V 1 . . . V 6 (af-ter Wagner and Altenberg 1996). There are more pleiotropic effectswithin the two modules than between them. Right: A path modelseparately showing common factors (G3, G5) that affect both mod-ules, and local factors (G1, G2, G4, G6) whose effects are confinedto only one module (after Mitteroecker and Bookstein 2007). Theleft and the right graphs have the same topology; the genes areonly differently arranged.

respond to selection on this part alone, independently of the rest ofthe phenotype, with few or no deleterious pleiotropic side effects.Conversely, traits with a common genetic or developmental basisare inherited jointly and evolve together (Lande 1979; Cheverud1996a). Thus, a modular organization enhances the ability of thegenetic system to generate adaptive variants, which is often re-ferred to as “evolvability” (Altenberg 1994, 2005).

The related concepts of correlation pleiades (Terentjev 1931)and morphological integration (Olson and Miller 1959) havebeen frequently applied in paleobiology and evolutionary biol-ogy. This literature emphasizes that traits affected by commongenetic, developmental, or environmental causes will appear to bephenotypically correlated within a species. For instance, a num-ber of morphometric studies reported tight statistical associationsamong measures of characteristics with a common developmen-tal origin (e.g., Zelditch 1987; Wagner 1990; Nemeschkal 1999;Klingenberg et al. 2003, 2004; Marroig et al. 2004; Hallgrımssonet al. 2007). In his theory of the “imitatory epigenotype,” Riedl(1978) further postulated that functionally related characters willevolve a common genetic basis. He argued that when parts of anorganism, for example, the elements of a bony joint, are func-tionally dependent, genetic and developmental integration amongthese parts may avoid deleterious independent variation and sofacilitate adaptive evolution. This notion is supported by quantita-tive genetic models (Cheverud 1984; Burger 1986; Wagner 1988)and also numerous empirical studies report concordance betweenfunctional relatedness among characters and their pattern of co-variation and coinheritance (e.g., Olson and Miller 1958; Atch-ley et al. 1981; Zelditch and Carmichael 1989; Cheverud 1995;Badyaev and Foresman 2004; Hallgrımsson et al. 2004).

Mitteroecker and Bookstein (2007) devised a formal modelof common and local developmental factors that differently affectphenotypic variables. Local factors contribute to morphological

variation within one module only, whereas common factors af-fect traits of different modules (Fig. 1, right). Modules are thusconstrued as anatomical parts (sets of measured morphometricvariables) that are influenced separately by dissociated local fac-tors and that may also be integrated through common develop-mental factors. The model specifically emphasizes the simultane-ous presence of modularity and developmental integration: somecharacteristics or shape features are independent across modules,whereas other properties are common to several modules—thatis, they are integrated. This factor model closely resembles thegenotype–phenotype map of Wagner and Altenberg. The genes intheir model (the Gs in Fig. 1, left) can be separated into those withpleiotropic effects on both modules (common factors) and thosewith effects on one module only (local factors), resulting in theright graph of Figure 1.

The factor model differs from the genotype–phenotype mapin that it is used to model the empirical covariance structure in thefollowing statistical analysis rather than any underlying biologicalprocesses. A “factor” is taken in its most general sense, includingnot only genes with various pleiotropic effects, but also geneticlinkage and epigenetic processes, such as chemical and mechan-ical tissue interactions during development (Muller and Newman2003). Unlike Wagner and Altenberg, we do not refer to module-specific functions here. The factor model focuses on the actualintegration and modularity observable on the morphological levelas caused by a series of developmental processes and genomicproperties, but it does not separate out these various causes.

Although our understanding of modularity is phrasedcausally in terms of developmental processes (in interaction withfunctional and selective regimes), the empirical approach to mor-phological integration is often based on observed correlationsamong morphological measures. Local genetic, developmental,and functional factors contribute to the covariances among phe-notypic variables within one module, but not to covariances be-tween modules. Hence, the assumption of many morphometricmethods for identifying modules is that covariances or correla-tions within one module are higher than covariances betweenmodules—groups of correlated characters (correlation pleiadesor ! -groups) are often interpreted as modules (see Chernoff andMagwene 1999 and Mitteroecker 2007 for review). Mitteroeckerand Bookstein (2007) showed that such expected covariances maybe confounded by the allometry of common and local factors andby factors with opposite effects on the same variables. The iden-tification of modules from observed covariances or correlationsalone requires very stringent and partly unrealistic assumptionsand is therefore unreliable.

Empirical models of integration and modularity are neverthe-less of direct relevance for phylogenetic reconstruction, estima-tion of selection gradients, and adaptive explanations of observedspecies differences (Gould and Lewontin 1979; Cheverud 1994;

944 EVOLUTION APRIL 2008

MODULARITY AND INTEGRATION IN THE HOMINOID CRANIUM

Wagner and Altenberg 1996; Strait 2001; Ackermann 2002). Thepresent article focuses on the hominoid cranium, which has under-gone considerable morphological change in the course of humanevolution. Numerous adaptive explanations have been devisedfor these alterations, typically assuming independent evolutionof body components (e.g., Ruff 1994; Churchill 1998; Aiello andDean 2002). Furthermore, most cladistic analyses treat their char-acters as mutually independent, providing separate evidence forevolutionary history. In contrast, empirical studies have reportedvarious levels of integration among different cranial elements andshape features during primate development and evolution (e.g.,Ross and Ravosa 1993; Ross and Henneberg 1995; Cheverud1995; Lieberman et al. 2000a,b; Marroig and Cheverud 2001;Strait 2001; Bookstein et al. 2003; Ackermann 2005; Bastir andRosas 2005, 2006; Gunz and Harvati 2007).

In the following analysis, we will specifically address themost prominent cranial modules, the face (viscerocranium orsplanchnocranium) and the cerebral capsule (neurocranium orneurobasicranial complex). We do not identify these modules frommorphometric variables, but infer them from known developmen-tal processes and functional properties of the cranium as brieflyreviewed here. It is well documented that the developing brain isspatially separated very early in embryogenesis from what willlater become the face and exhibits a different pre- and postnatalgrowth pattern. Already after the neural folds fuse to form theneural tube, the neuroectoderm and the surface ectoderm are twodiscrete tissues. At the end of the first month, the human faceis represented by the stomodeum (the future oral cavity) and thefirst pharyngeal arch, which develop largely independently of theneural tube giving rise to the brain (see, e.g., Sperber 2001; Helmset al. 2005; Tapadia et al. 2005).

During the formation of the cranial skeleton, all the bone thatcontributes to the viscerocranium derives from neural crest cells,whereas large parts of the cranial vault are of mesodermal origin.The bones of the face and the sides and the roof of the neuro-cranium are formed by intramembranous (dermal) ossification,whereas the cranial base (parts of the occipital, sphenoid, tem-poral, ethmoid) is mainly formed by endochondral ossification.The occipital region surrounding the foramen magnum derivesfrom the sclerotome of the occipital somites and hence differsfrom the rest of the skull in both its tissue origin and its modeof ossification (Noden 1988; Lieberman et al. 2000b; Morris-Kay2001; Jiang et al. 2002). During postnatal development, the neu-rocranium and the orbital cavities follow the typical neural growthpattern, characterized by rapid early size increase ceasing around5 to 7 years of age (in humans). But most of the viscerocraniumfollows the general growth pattern of somatic tissue, increasing insize right up through early adulthood (Enlow and Hans 1996).

Despite the early spatial separation and differentiation of thedeveloping face from the brain and its surrounding cranial cap-

sule, a number of growth factors and morphogens with pleiotropic(and hence integrating) effects on several parts of the vertebratecranium do exist. Examples of important and well-documentedpleiotropic developmental genes and gene families are SHH, FGF,BMP, Retinoic acid, and the postnatal GH/IGF axis (for recentreviews see, e.g., Hu and Helms 1999; Helms et al. 2005; Tapadiaet al. 2005; Funatsu et al. 2006; Nie et al. 2006a,b). Gross pheno-typic effects of such factors (as observed mainly from knockoutexperiments) are remarkably similar across organisms as differ-ent as zebrafish, chicken, mice, and humans, and the correspond-ing signaling pathways are largely conserved during evolution.Tissue-specific expression patterns and distributions of receptors,as well as stage-dependent properties relying on surrounding tis-sues for inductive influences, often contribute to modular effectsof those factors at different sites and at different developmentalstages. Conversely, variation in the expression of the genes or theirreceptors during early embryogenesis along with allelic variationcan lead to integrated effects throughout the whole head.

Furthermore, tissues always interact with other tissues as theygrow—the induced mechanical forces are decisive factors in cre-ating organic form (Herring 1993; Blechschmidt and Freeman2004; Radlanski and Renz 2006). Regardless of any genetic con-tingency, the necessity of maintaining a spatially and functionallyintegrated cranium contributes to the epigenetic integration be-tween the face and the neurocranium (Weidenreich 1941; Mossand Young 1960; Enlow et al. 1969; Roth 1996). For example,studies of artificial deformations of the human cranial vault knownfrom various cultures showed that facial and basicranial shape aresignificantly affected by the mechanical forces induced (Anton1989; Cheverud et al. 1992; Kohn et al. 1993; Rhode and Arriaza2006). Even beyond the period of neurocranial growth, mastica-tory muscles provide a further source of epigenetic integrationbetween the face and the neurocranium. For instance, the inser-tions of the temporal muscles cause superficial modifications ofthe vault that may become massive sagittal crests as seen in gorillas(Washburn 1947; Riesenfeld 1955; Robinson 1958).

RESEARCH QUESTIONS AND HYPOTHESES

Based on this extensive embryological and experimental evidencefor modularity of the primate face and neurocranium as well as fortheir integration, we statistically estimate aspects of shape varia-tion that are common to the hominoid face and the neurocranium(integrating common factors) and also the aspects of shape varia-tion due to modular developmental factors (local factors affectingeither the face or the neurocranium) from measured shape coor-dinates on dried crania. These factor estimates are then used todescribe evolutionary shape changes in terms of integrated andmodular developmental variation.

Because different modules often serve different functions,specific selection regimes likely involve only one or a few

EVOLUTION APRIL 2008 945


modules at a time. Consequently, pleiotropic effects may inter-fere with adaptation due to deleterious side effects in the othermodules, so that common factors are expected to be relativelyconserved during evolution. Local factors are less constrained andmay respond more easily to varying selection pressures. This ex-pectation seems to be supported by the fact that highly pleiotropicdevelopmental genes have remarkably similar functions duringcranial development in a wide range of vertebrate species (e.g.,Morris-Kay 2001; Santagati and Rijl 2003; Hanken and Gross2005; Helms et al. 2005). Furthermore, Khaitovich et al. (2005)found that expression levels as well as DNA sequences of genesactive in more tissues have diverged less between humans andchimpanzees than have genes active in fewer tissues.

We therefore expect (1) that common factors are conservedacross hominoid species, which is to say, they have similar effectsin all three species and contribute little to their shape differences;and (2) that most evolutionary shape differences mainly owe toalterations of local factors. Any break-up of ancestral patternsof developmental integration in the course of evolving the ac-tual species differences of mean shape would implicate differentmodule-specific regimes of directional and stabilizing selectionand so would suggest speculations about certain evolutionary sce-narios (Wagner and Altenberg 1996; Cheverud 1996a; Wagneret al. 2005). According to the definition of a factor, developmen-tal integration is taken here in a wide sense as due to genetic andepigenetic factors during development along with genetic link-age. In the absence of directed selective forces, within-speciesand between-species covariance matrices should be proportional(Lande 1979) and developmental integration would be expectedto resemble evolutionary integration (covariation across the threespecies means).

Materials and MethodsThree-dimensional coordinates of 45 paired and unpaired anatom-ical landmarks and of 157 semilandmarks on ridge curves andthe neurocranial surface (Fig. 2) were measured with a Micro-

Figure 2. A juvenile cranium of a chimpanzee with the 347 land-marks and semilandmarks. The gray landmarks belong to the faceand the black ones to the neurocranium.

scribe 3DX digitizer (Immersion Corporation, San Jose, CA) ona cross-sectional sample of dried crania. The sample comprisesspecimens of both sexes from three closely related but morpho-logically diverse genera: 52 Homo sapiens, 49 Pan troglodytes,and 56 Gorilla gorilla. The age of the specimens ranges fromnewborns to adults in all three species. The landmarks cover allexternal parts of the face, the cranial vault, and the midline cranialbase. Semilandmarks were measured on one side of the craniumonly, but to enable three-dimensional surface representations theywere warped and reflected with a thin-plate spline algorithm ontothe other side based on the measured anatomical landmarks. Formore technical details see Gunz (2005) and Mitteroecker (2007),and for information on the sample see Bernhard (2003), and Mit-teroecker et al. (2004). Semilandmarks are points sampled alongoutlines or surfaces that are allowed to slide along their curvatureso as to minimize “bending energy,” a quantity measuring localshape differences versus the mean shape. Semilandmarks can beused in the subsequent analytic toolkit of geometric morphomet-rics as if they were homologous point locations (Bookstein 1997;Gunz et al. 2005).

The 157 sets of 347 landmarks and semilandmarks weretransformed into shape coordinates by one overall GeneralizedProcrustes Analysis (Rohlf and Slice 1990). Thereafter, theywere separated into two modules (Fig. 2): the face with 146(semi)landmarks and the neurocranium with 201 (semi)landmarks(see also the Appendix on alternative analyses). All morphometricand statistical analyses were performed in Mathematica 5.2 basedon routines written by Philipp Mitteroecker and Philipp Gunz.Surface representations were rendered in Amira 3.0.

ASSESSMENT OF DEVELOPMENTAL INTEGRATION

To estimate common factors, that is, dimensions of shape varia-tion that are integrated among the face and the neurocranium, weemploy the two-block partial least squares (PLS) approach thatis called singular warp (SW) analysis when applied to Procrustescoordinates (Bookstein et al. 1996, 2003; Rohlf and Corti 2000).For each extracted dimension, the analysis yields two singularvectors, one for the face and one for the neurocranium, that can beconstrued as the two shape changes that most highly covary in thesample. In Mitteroecker and Bookstein (2007) we demonstratedthat the two PLS loading vectors serve together as a common fac-tor estimate when the vectors are scaled appropriately so that theyrelate to common factors in Sewall Wright’s (1932) factor analysisapproach (see Appendix for algebraic details).

Based on these common factor estimates, the full shape spaceis partitioned into three subspaces: an integrated shape spacespanned by the common factors, and two modular shape spaces,one for the face and one for the neurocranium, that complementthose dimensions of integrated shape variation. The integratedspace is ordinated by scores along the common factors, and the



modular spaces by scores along principal component axes of theresidual data after the common factors have been removed fromboth blocks of variables separately. These scores let one expressthe observed shape variation in terms of common and local de-velopmental processes. By design, the integrated shape space ac-counts for shape differences due only to common factors, whereasshape differences in the modular spaces are due mainly to localfactors. To evaluate the relative contribution of integrated ver-sus modular shape variation to the observed shape differences,specimens and ontogenetic trajectories are compared in all threesubspaces (O’Higgins 2000; Mitteroecker et al. 2004, 2005). Forexample, if the species completely overlap in the integrated spacebut differ in the modular spaces, one would conclude that evolu-tionary shape differences among the species are due to alterationsin modular developmental processes.

When relating evolutionary differences to developmental in-tegration, the choice of the “reference” sample whose covariancesdrive the PLS is crucial. In analyses pooling all the adult speci-mens of the three species, most of the variation and covariationin shape is due to the mean species differences. Accordingly, thefirst dimensions of PLS would describe evolutionary integration—how the face and the neurocranium covary across the three speciesmeans—which is not the same as developmental integration. Al-though developmental integration is due to common genetic andepigenetic effects, evolutionary integration is a result of develop-mental integration and coinheritance in the context of selectiveregimes. Similarly, PLS based on the full cross-sectional sam-ple including subadult specimen would mainly assess covariationacross average postnatal age stages. The appropriate PLS anal-ysis to assess developmental integration has to be based on thecovariation among adults within one species, which is the resultof common and local morphogenetic factors during the full pe-riod of pre- and postnatal development. For the present analysis ofthree different species, PLS is based on the pooled within-speciescovariance matrix. See also the Appendix for alternative analyses.

Differences of integration among species as well as differ-ences between developmental integration and evolutionary inte-gration can be identified in plots of the SW scores for the faceagainst those for the neurocranium, one plot for each paired di-mension (common factor). To the extent that patterns of develop-mental integration are identical among all species and also iden-tical to evolutionary integration, all specimens would lie close toa straight line in these plots.

ResultsWe performed a two-block PLS analysis on the adult pooledwithin-species covariance matrix and extracted four common fac-tors. Table 1 shows that nearly 90% of the squared covariance (inthe pooled adult sample) is already explained by the first two di-mensions of PLS, but a permutation test (see Appendix) yields four

Table 1. For the first four extracted dimensions of the singularwarp (SW) analysis, this table provides the covariance (in unitsof squared Procrustes distance !104) and correlation among thesingular warp scores of the two blocks, the percentage of the adultwithin-species variance explained by the common factors (scaledPLS loading vectors), the percentage of the full variance explainedby the common factors, and the percentage of the adult within-species covariance explained by the common factors.

SW Covar. Correl. Expl. var. Expl. Expl.pooled var. full covar.

1 4.25 0.85 32.9% 44.1% 71.2%2 2.13 0.82 12.3% 30.6% 17.9%3 0.97 0.75 5.9% 1.8% 3.7%4 0.83 0.77 5.1% 2.4% 2.7%

dimensions that differ significantly from a random distribution.Correlation as well as covariance among PLS scores decreasesmarkedly after the fourth factor.

Figure 3 shows the space of integrated shape variation asindividual scores along the common factors. The ontogenetic tra-jectories of Pan and Gorilla overlap in these four dimensions,where Gorilla clearly extends the common trajectory. The shorthuman trajectory differs somewhat in direction from that of theapes, indicating less and slightly different postnatal shape change.

The shape deformations depicted by the common factors arevisualized in Figures 4 and 5 as landmark displacement vectorsand as deformations of a surface representation of a chimpanzeecranium extracted from a CT scan. The shape changes visualizedcorrespond to increasing scores in Figure 3. The first common fac-tor involves an enlargement and forward protrusion of the maxillaalong with a relatively smaller cranial vault with both sagittal andnuchal crests. High scores along the second common factor corre-spond to a broad and large face and a broad but short neurocranium,whereas low scores are associated with long and narrow crania.The third factor involves a reorientation of the alveolar process anda shape change of the occipital. The fourth common factor is a con-trast between high and more spherical versus more flat and ellip-soidal neurocrania, where the latter, elongated, shape is associatedwith pronounced supraorbital tori and enlarged zygomatic arches.

Figures 6 and 7 show principal component scores of the mod-ular shape spaces that are devoid of those dimensions of inte-grated shape variation. Shape differences depicted by these twofigures reflect differences due to local factors, which is to say,modular shape variation. In the modular space for the face, thethree species-specific trajectories diverge during postnatal devel-opment, and human development is clearly distinct by the time ofbirth. In the neurocranial modular space, chimpanzee and gorillatrajectories overlap in the first four PCs but differ along the fifthPC. The gorilla trajectory is again much longer than Pan’s, andthe human trajectory deviates from both of them.



Figure 3. The integrated shape space reflects cranial shape variation due to the four common factors—variation that affects both theface and the neurocranium. Adult and subadult specimens (larger and smaller points, respectively) for all three species are shown in thisspace. The developmental sequence of specimens from one species traces out an ontogenetic trajectory. The approximate geometry ofthe three trajectories is schematized in Figure 9.

Figure 8 shows all four dimensions of singular warp scoresfor the face plotted against the scores for the neurocranium. Theseare scores along the common factors, computed separately forthe face and the neurocranium, to assess the actual integration ofthe corresponding shape features due to the common factors. Theorientation of the adult specimens’ point clouds is very similaramong the three species, especially for the first two dimensions ofPLS, illustrating similarity (i.e., conservation) of developmentalintegration patterns. That is, one unit of shape change along theneurocranial component is associated with an amount of facialshape change that is similar in all three species, irrespective of theactual average shape of the species. In the first two dimensions, alladult and subadult specimens lie along one line, suggesting thatthe species differences are also integrated in that way. In the thirdand fourth dimensions, integration in humans differs slightly fromthe others. Also, the human group is displaced as a whole from thediagonal, indicating that the way humans differ from chimpanzeesand gorillas is not integrated in these dimensions.

DiscussionINTEGRATION DURING CRANIAL DEVELOPMENT

There is extensive embryological and experimental evidence forthe presence of common developmental factors or processes (thataffect both the neurocranium and the face) as well as for localfactors (affecting either the face or the neurocranium) during pri-mate craniofacial development. It is thus beyond question that theprimate face and neurocranium possess some shape characteris-tics that are modular and others that are integrated among the tworegions. Even though humans differ considerably in average cra-

nial shape and particularly in their postnatal growth pattern fromother primates (e.g., Cobb and O’Higgins 2004; Mitteroecker et al.2004), the major developmental processes are certainly sharedamong all hominoids (as they are conserved even across a muchwider taxonomic range; see, e.g., Morris-Kay 2001; Helms et al.2005; Tapadia et al. 2005). Accordingly, we find that Homo, Pan,and Gorilla have very similar but not quite identical patterns ofdevelopmental integration (similar slopes within adults in Fig. 8),implying that the effects of common factors are relatively con-served among hominoids. These results support those of Acker-mann (2002, 2005) and are in accordance with findings that cranialintegration is similar across a broad taxonomic range of primates(e.g., Cheverud 1996b; Marroig and Cheverud 2001; Marroig et al.2004) and even therian mammals (Goswami 2006). Claims byPolanski and Franciscus (2006) that the hominoid face is “un-coupled” from the neurocranium and that humans, in contrast toapes, are not integrated in the face is likely a consequence of theirproblematic methodology based on the inverse covariance matrix(cf. Mitteroecker and Bookstein 2007). In fact, a series of recentstudies described tightly integrated shape features in the humancranium (Lieberman et al. 2000a; Bookstein et al. 2003; Gonzalez-Jose et al. 2004; Ackermann 2005; Bastir and Rosas 2005, 2006;Bastir et al. 2006).

To show the actual patterns of integration, Figures 4 and 5visualize the four pairs of SWs as common factors, that is, asjoint shape deformations. The first factor is mainly related tomastication—an enlarged and prognathic maxilla along with arelatively small cranial capsule is associated with cranial crestsand enlarged zygomatic arches. These characteristics are epigenet-ically associated in primates mainly through masticatory muscle



Figure 4. Visualization of the first two common factors as shape deformations. The first column of images describes the accordingshape changes by landmark displacement vectors (from the consensus). Columns 2 to 5 are corresponding surface morphs where eachmorph differs from its neighbors by 1.5 standard deviations of the actual variability in the adult sample. The depicted shape changes(the sequence from the left morphs to the right ones) correspond to increasing scores in Figure 3. Each morph and also the landmarkdisplacement vectors are shown in a frontal, a lateral, and a superior view (top, middle, and bottom images). Because no landmarks weredigitized on the teeth, the dentition is not involved in the warping.

activity. The second factor contrasts broad and short crania withnarrow and long crania, including both the face and the neuro-cranium (brachycephalic versus dolichocephalic crania), and alsoinvolves changes in the overall size of the face relative to theneurocranium. This somewhat uniform cranial shape deformationis a commonplace finding in cephalometrics and has also beenidentified in previous studies of morphological integration (e.g.,Weidenreich 1941; Enlow and Hans 1996; Lieberman et al. 2000a;Bookstein et al. 2003; Bastir and Rosas 2004). Interestingly, thedimensions of the nasal aperture are largely unaffected by theseoverall shape changes, corroborating results from Anton (1989)and Rhode and Arriaza (2006) that artificial deformations of the

cranial vault do not affect the nasal aperture but only more periph-eral facial structures. Shape change along the third common factorencompasses relative size of the midface and neurocranial globu-larity, two characteristics that are tightly associated during postna-tal ontogeny. Furthermore, the positions of the foramen magnumand the superior nuchal line are related to facial kyphosis, per-haps as a joint association with locomotion and posture behavior(Schultz 1942; Manfreda et al. 2006) or with the basicranial angleas driven by relative brain size (Ross and Ravosa 1993; Liebermanet al. 2000a,b). The fourth common factor contrasts crania with aroundish and relatively short and high neurocranium to elongated,ellipsoidal crania. Exaggeration of the latter generates an occipital



Figure 5. Visualization of the third and fourth common factors as shape deformations.

bun and lambdoid flattening along with large browridges in theupper face—a combination of traits that is quite characteristicfor archaic human morphology (e.g., Trinkaus and LeMay 1982;Lieberman et al. 2002; Schwartz and Tattersall 2002, 2003; Gunzand Harvati 2007). These four described factors do not necessar-ily directly relate to four single biological causes. Actual shapedifferences and biological processes may instead consist of linearcombinations of these four factors. Also, as the shape character-istics we have been describing change jointly during postnataldevelopment, the factors are necessarily highly correlated overontogeny (Fig. 3).

THE DEVELOPMENTAL BASIS OF EVOLUTIONARY

CHANGE

Evolutionary theory leads one to expect that characteristics in-duced by local developmental factors are less constrained in theirevolution than such properties caused by common factors. We

therefore expected that local factors would contribute most tothe average cranial shape differences among species. As sum-marized in Figure 9, variation in both common and local factorscontributes to the cranial shape differences among humans, chim-panzees, and gorillas, but contrary to our expectation, the firstfour common factors explain 79% of shape variation of the fulldata and 83% of variation among the adult species mean shapes.These numbers are only of limited biological relevance as theydepend on the distribution of landmarks and other details. Still,they may indicate that cranial evolution in hominoids has to a largeextent been achieved by alterations of common growth factors—mutations of genes with numerous pleiotropic effects. These find-ings corroborate one earlier speculation of King and Wilson (1975)that only few mutations of regulatory genes—genes with poten-tially many pleiotropic effects—may account for the pronouncedmorphological differences between humans and chimpanzeesdespite their close phylogenetic and molecular proximity (see,



Figure 6. The modular shape space of the facial landmarks reflectsshape variation due to local developmental factors in the face. Thefigure shows the first three principal component (PC) scores of thisspace. The legend of this and the next two figures is as in Figure3. In contrast to the integrated shape space, all three ontogenetictrajectories are clearly distinct here.

e.g., Mann and Weiss 1996; Ruvolo 1997; Mitteroecker et al.2004).

Figure 9A schematizes that the ontogenetic trajectories inthe integrated shape space, that is, postnatal shape changes dueto common developmental factors, are in a common directionfor all three species. Chimpanzee and gorilla trajectories overlapcompletely in this space, where adult chimpanzees overlap withsubadult gorillas, and humans diverge only slightly from them.But in the modular components of shape (Fig. 9B,C), trajectoriesare all different and humans diverge markedly in the face as wellas in the neurocranium. These results suggest that common fac-tors contribute considerably to evolutionary shape differences, butmainly by extension or truncation of otherwise conserved devel-opmental pathways—an evolutionary phenomenon that is oftencalled heterochrony or ontogenetic scaling (Gould 1977; Zelditch2001; Mitteroecker et al. 2005). The analysis confirms that localfactors and hence modular shape characteristics possess more de-grees of freedom for evolutionary change than mere ontogeneticscaling.

EVOLUTIONARY INTEGRATION AND MODULARITY

Figure 8 plots the SW scores for the face versus the scores for theneurocranium—scores for shape characteristics that are tightly

Figure 7. The modular shape space of the neurocranial landmarksreflects shape variation due to local factors in to the neurocranium.The human ontogenetic trajectory is clearly distinct whereas thetrajectories of chimpanzee and gorilla largely overlap in the firstfour PCs and differ only along the fifth PC.

integrated by common developmental factors. The association offacial and neurocranial scores among adults is similar within allthree species for all four dimensions, indicating similar patternsof integration. In the first two dimensions, the specimens of allthree species closely scatter around the diagonal. This single tra-jectory shows that evolutionary shape differences along these twocomponents follow the common pattern of developmental integra-tion. Along the third and also the fourth dimension, in contrast,humans deviate markedly from the common nonhuman trajec-tory. For these shape features, similarly integrated within all threespecies, humans differ from nonhumans in a nonintegrated way.Some aspects of human cranial shape thus have evolved in anintegrated fashion (common factors 1 and 2) whereas other char-acteristics have been decoupled during human evolution (commonfactors 3 and 4). In contrast, evolutionary integration between Panand Gorilla closely reflects developmental integration in all fourof these dimensions.

It is unlikely that the evolutionary dissociation has beenachieved through alterations of developmental integration, that is,of the differential effects of common factors, because we foundthat developmental integration in recent humans is relatively sim-ilar to that of chimpanzees and gorillas. Also, the physical causesunderlying epigenetic integration may not be subject to evolu-tionary change. A more likely explanation for the evolutionarydissociation is that integrated aspects of shape—significantly con-trolled by common factors—may additionally be affected by morelocal factors. This means not only that different aspects of shapeare controlled by different local and common factors, but also that



Figure 8. Singular warp (SW) scores for the face against the scores for the neurocranium. They show the actual association of facial andneurocranial shape characteristics due to the four common factors. The axes are not scaled isometrically as their scales also depend onthe number of landmarks and are thus not meaningful.

the same shape characteristics are also controlled simultaneouslyby local and common factors. Such overlapping morphogeneticcontrol would allow independent response to selection for traitsthat are also integrated due to common genetic and developmentalfactors. (Compare also Hansen [2003], who argued that a combi-nation of local and common factors optimizes evolvability whenassuming a constant total variance.) The idea of redundant mor-phogenetic factors is related to the phenomenon of gene fami-lies consisting of numerous members with related functions butvarying pleiotropic effects. The gene families involved in cran-iofacial development include many members (compare, e.g., theBMP gene family, Nie et al. 2006a, p. 513); even most of the cor-responding receptor genes have multiple paralogs (see, e.g., themultiple roles of FGF receptor genes in causing craniosynostosis:Aleck 2004; Marie et al. 2005).

To measure the redundancy of common and local factors wecomputed the fractions of variance along the four SWs that re-main unexplained by the common factors (pooled for all threespecies). These fractions depict the modular variance of char-

acteristics that are simultaneously under strong common factorcontrol. Along the first common factor there is considerable ad-ditional modular variance for the face (0.204) but nearly none forthe neurocranium (0.007). A number of factors, including localones, influence the outgrowth of the face and the prominence ofthe masticatory apparatus, whereas cranial crests (the shape fea-tures influenced by common factor 1, see Fig. 4) are largely apostnatal response to masticatory muscles, with no local factorsin the neurocranium involved (Washburn 1947; Riesenfeld 1955;Robinson 1958). For common factor 2—the overall dimensionsof face and neurocranium—the face exhibits a higher modularvariance (0.118) than the neurocranium (0.065), perhaps becausethe dimensions of the cranial vault strongly influence the facebut not vice versa (Weidenreich 1941; Anton 1989; Lieberman2000a). Along factor 3, there is more unexplained variance in theneurocranium (0.172) than in the face (0.081). The correspond-ing neurocranial shape change mainly involves the nuchal planeand the positions of the foramen magnum and of the superiornuchal line. In addition to common factors, those characteristics



A

B C

Modular shape space: Face Modular shape space: Neurocranium

Integrated shape space

Homo

Pan

Gorilla

Figure 9. Schematization of the geometry of the three ontogenetic trajectories in (A) the integrated shape space (see Fig. 3), (B) themodular shape space of the face (Fig. 6), and (C) the modular shape space of the neurocranium (Fig. 7). Integrated shape variation amongthe three species is largely constrained along one single trajectory whereas the development of modular shape characteristics variesmore fundamentally.

also depend on relative brain size, body posture, and locomotion,as well as on the prominence of the nuchal muscles—factors thatcertainly affect the neurocranium more directly than the face andso contribute to local variance. Additional local variance alongcommon factor 4 exists mainly for the face (0.186) and less forthe neurocranium (0.057). Corresponding facial shape change in-volves mainly the browridges that start to develop when the brainhas already ceased to grow and the overall dimensions of the neu-rocranium are established. There is necessarily independent localvariance present in those aspects of the face.

We thus find considerable modular variance for several in-tegrated traits in accord with our experimental understanding ofcranial development. Overlap of common and local developmentalcontrol for the same shape characteristics appears as a reasonablemodel for dissociated human cranial evolution. It is further ev-ident from Figure 3 that humans exhibit less variance along thefour common factors than do chimpanzees and gorillas. In fact,modular factors contribute considerably more to human adult vari-ation (65%) than to chimpanzees (48%) and gorillas (34%). Hu-mans thus share very similar patterns of covariation between faceand neurocranium, but the underlying common factors contributeless to adult shape variation. There is relatively more variation inmodular processes during human development as compared to thedevelopment of chimpanzees and gorillas.

It is not clear whether all hominoids share the same patternof genetic redundancy among common and local processes andmay thus potentially have the same ability for local responsesto selection, or whether only humans evolved additional local

morphogenetic control. In the first case, some aspects of hu-man morphology would owe to evolutionary change of local fac-tors that are also present in other hominoids. These local fac-tors need not contribute much to adult shape variation in allspecies. Perhaps constrained by stabilizing selection or canal-izing properties, they may rather represent some “latent modu-larity.” Such relationships have been demonstrated recently fordifferent traits in butterfly morphology (Beldade et al. 2002;Frankino et al. 2005). Traits that exhibit tight genetic and pheno-typic correlations responded independently to artificial selectionbut maintained their correlations within the differently selectedpopulations.

To address this question, we computed the modular variancesalong common factor 3 and 4 for each species separately. Humanshave a modular variance along common factor 4 (face 0.191/neu-rocranium 0.047) that is comparable to that of nonhumans (chimps0.139/0.049; gorillas 0.219/0.077) so that all three species mightshare the same local factors and hence the same pattern of modu-larity. This may reflect the fact that browridge morphology—themajor effect of common factor 4 on the face—is largely deter-mined after the neurocranium has already ceased to grow, and isthus similarly independent of neurocranial morphology in all threespecies. For common factor 3, in contrast, humans possess by farthe most modular variance in both the face and the neurocranium(humans 0.281/0.287; chimps 0.037/0.107; gorilla 0.079/0.165).This may be indicative of a larger number of polygenic loci (cf.Burger and Lande 1994) and thus of a (relatively) increased ex-tent of local factors in the development of these shape features



in humans. But it could also result from the distinctive humandevelopmental pathway itself, for example, from increased activ-ity of these local factors during development.

Although the evolutionary dissociation of humans along com-mon factor 4 likely owes to local factors that are present in allthree species, dissociation along common factor 3 might have re-quired the emergence of novel local developmental control mech-anisms in humans. Altenberg (1994) regarded “constructional se-lection,” the evolutionary emergence of new local factors, as themajor mode of the evolution of modularity. This closely relates togene duplication and divergence during evolution, increasing thenumber of genes with related effects that yet may differ in theirpleiotropic range. Gene duplication is regarded as a major drivingforce of evolutionary change (e.g., Ohno 1970; Zhang 2003) and isinvolved in a number of aspects of primate and human evolution(Gagneux and Varki 2001; Zhang 2003), including craniofacialmorphology (Fortna et al. 2004; Cheng et al. 2005). Approxi-mately a third of the human genes are duplications (Zhang 2003)and about 33% of the human duplications are not duplicated inchimpanzees (Cheng et al. 2005); most of these differences aredue to novel duplications during human evolution.

In summary, morphological differences among humans,chimpanzees, and gorillas have been achieved by evolutionarychanges of both common and local developmental factors. Fac-tors with pleiotropic effects on both the face and the neurocra-nium account for a large amount of interspecific variation buttheir differential effects on cranial shape appear relatively con-served during hominoid evolution. Local factors (i.e., modularshape characteristics), in contrast, possess more degrees of free-dom for evolutionary change. However, some aspects of facialand neurocranial morphology that are tightly integrated duringdevelopment evolved in a dissociated way to bring about humanmorphology. This dissociation has perhaps been achieved by evo-lutionary alterations of local factors affecting characteristics thatare simultaneously controlled by common factors. For some as-pects of shape, the dissociation might have required the emergenceof novel local factors during human evolution.

ACKNOWLEDGMENTSWe are grateful to W. Callebaut, P. Gunz, S. Katina, B. Metscher, K.Schaefer, G. Weber, and three anonymous reviewers for thoughtful com-ments on earlier versions of this manuscript and to H. Seidler, K. Schaefer,and G. Weber for their support during this work. We thank M. Bernhardfor sharing the data that he collected in the course of his Ph.D. research,P. Gunz for the extensive data preparation, and the curators and staff ofthe museums and departments in Belgium, Switzerland, Germany, andAustria, who allowed access to the collections in their care. This studywas supported by grant GZ 200.033/1-VI/I/2004 of the Austrian Councilfor Science and Technology to H. Seidler, EU FP6 Marie Curie ResearchTraining Network MRTN-CT-019564, and the Konrad Lorenz Institutefor Evolution and Cognition Research, Altenberg, Austria.

LITERATURE CITEDAckermann, R. R. 2002. Patterns of covariation in the hominoid craniofacial

skeleton: implications for paleoanthropological models. J. Hum. Evol.43:167–187.

———. 2005. Ontogenetic integration in the hominoid face. J. Hum. Evol.48:175–197.

Aiello, L., and C. Dean. 2002. An introduction to human evolutionary anatomy.Academic Press, London.

Aleck, K. 2004. Craniosynostosis syndromes in the genomic era. Semin. Pe-diatr. Neurol. 11:256–261.

Altenberg, L. 1994. The evolution of evolvability in genetic programming.Pp. 47–74 in K. E. Kinnear, ed. Advances in genetic programming. MITPress, Cambridge, MA.

———. 2005. Modularity in evolution: some low-level questions. Pp. 99–128in W. Callebaut and D. Rasskin-Gutman, eds. Modularity: understandingthe development and evolution of complex natural systems. MIT Press,Cambridge, MA.

Anton, S. C. 1989. Intentional cranial vault deformation and induced changesof the cranial base and face. Am. J. Phys. Anthropol. 79:253–267.

Atchley, W., J. Rutledge, and D. Cowley. 1981. Genetic components of sizeand shape. II. Multivariate covariance patterns in the rat and mouse skull.Evolution 35:1037–1055.

Badyaev, A. V., and K. R. Foresman. 2004. Evolution of morphological inte-gration: I. Functional units channel stress-induced variation. Am. Nat.163:868–879.

Bastir, M., and A. Rosas. 2004. Facial heights: evolutionary relevance of post-natal ontogeny for facial orientation and skull morphology in humansand chimpanzees. Am. J. Phys. Anthropol. 47:359–381.

———. 2005. Hierarchical nature of morphological integration and modular-ity in the human posterior face. Am. J. Phys. Anthropol. 128:26–34.

———. 2006. Correlated variation between the lateral basicranium and theface: a geometric morphometric study in different human groups. Arch.Oral. Biol. 51:814–824.

Bastir, M., A. Rosas, and P. O’Higgins. 2006. Craniofacial levels and themorphological maturation of the human skull. J. Anat. 209:637–654.

Beldade, P., K. Koops, and P. M. Brakefield. 2001. Developmental constraintsversus flexibility in morphological evolution. Nature 416:844–847.

Bernhard, M. 2003. Sexual dimorphism in the craniofacial morphology ofextant hominoids. Ph.D. Thesis, Department of Anthropology, Univ. ofVienna, Vienna.

Blechschmidt, E., and B. Freeman. 2004. The ontogenetic basis of humananatomy: the biodynamic approach to development from conception toadulthood. North Atlantic Books, Berkely, CA.

Bolker, J. A. 2000. Modularity in development and why it matters in Evo-Devo.Am. Zool. 40:770–776.

Bookstein, F. 1997. Landmark methods for forms without landmarks: mor-phometrics of group differences in outline shape. Med. Image Anal.1:225–243.

Bookstein, F., P. D. Sampson, A. P. Streissguth, and H. M. Barr. 1996. Ex-ploiting redundant measurement of dose and developmental outcome:new methods from the behavioral teratology of alcohol. Dev. Psychol.32:404–415.

Bookstein, F. L., P. Gunz, P. Mitteroecker, H. Prossinger, K. Schaefer, and H.Seidler. 2003. Cranial integration in Homo: singular warps analysis of themidsagittal plane in ontogeny and evolution. J. Hum. Evol. 44:167–187.

Burger, R. 1986. Constraints for the evolution of functionally coupled charac-ters: a nonlinear analysis of a phenotypic model. Evolution 40:182–193.

Burger, R., and R. Lande. 1994. On the distribution of the mean and varianceof a quantitative trait under mutation-selection-drift balance. Genetics138:901–912.

Callebaut, W., and D. Rasskin-Gutman. 2005. Modularity: understanding the



development and evolution of natural complex systems. Vienna Seriesin Theoretical Biology. MIT Press, Cambridge, MA.

Cheng, Z., M. Ventura, X. She, P. Khaitovich, T. Graves, K. Osoegawa, D.Church, P. DeJong, R. K. Wilson, S. Paabo, et al. 2005. A genome-widecomparison of recent chimpanzee and human segmental duplications.Nature 437:88–93.

Chernoff, B., and P. M. Magwene. 1999. Morphological integration: fortyyears later. Pp. 319–354. Morphological Integration. Univ. of ChicagoPress, Chicago.

Cheverud, J. M. 1984. Quantitative genetic and developmental constraints onevolution by selection. J. Theor. Biol. 110:155–171.

———. 1995. Morphological integration in the saddle-back tamarin (Saguinusfuscicollis) cranium. Am. Nat. 145:63–89.

———. 1996a. Developmental integration and the evolution of pleiotropy.Am. Zool. 36:44–50.

———. 1996b. Quantitative genetic analysis of cranial morphology in thecotton-top (Saguinus oedipus) and saddle-back (S. fuscicollis) tamarins.J. Evol. Biol. 9:5–42.

Cheverud, J. M., L. A. P. Kohn, L. W. Konigsberg, and S. R. Leigh. 1992.Effects of fronto-occipital cranial vault modification on the cranial baseand face. Am. J. Phys. Anthropol. 88:323–345.

Churchill, S. E. 1998. Cold adaptation, heterochrony, and neandertals. Evol.Anthropol. 7:46–61.

Cobb, S., and P. O’Higgins. 2004. Hominins do not share a common postna-tal facial ontogenetic shape trajectory. J. Exp. Zool. (Mol. Dev. Evol.)302B:302–321.

Enlow, D. H., R. E. Moyers, W. S. Hunter, and J. A. McNamara, Jr. 1969.A procedure for the analysis of intrinsic facial form and growth. Am. J.Orthod. 56:6–23.

Enlow, D., and M. Hans. 1996. Essentials of facial growth. Saunders Company,Philadelphia PA.

Fortna, A., Y. Kim, E. MacLaren, K. Marshall, G. Hahn, L. Meltesen, M. Bren-ton, R. Hink, S. Burgers, T. Hernandez-Boussard, et al. 2004. Lineage-specific gene duplication and loss in human and great ape evolution.PLoS Biol. 2:937–954.

Frankino, W. A., B. J. Zwaan, D. L. Stern, and P. M. Brakefield. 2005. Naturalselection and developmental constraints in the evolution of allometries.Science 307:718–720.

Funatsu, M., K. Sato, and H. Mitani. 2006. Effects of growth hormone oncraniofacial growth. Angle. Orthod. 76:970–977.

Gagneux, P., and A. Varki. 2001. Genetic differences between humans andgreat apes. Mol. Phylogenet Evol. 18:2–13.

Gonzalez-Jose, R., S. Van der Molen, E. Gonzalez-Perez, and M. Hernandez.2004. Patterns of phenotypic covariation and correlation in modern hu-mans as viewed from morphological integration. Am. J. Phys. Anthropol.123:69–77.

Goswami, A. 2006. Cranial modularity shifts during mammalian evolution.Am. Nat. 168:270–280.

Gould, S. J. 1977. Ontogeny and phylogeny. Harvard Univ. Press, Cambridge.Gunz, P. 2005. Statistical and geometric reconstruction of hominid crania:

reconstructing australopithecine ontogeny. Ph.D. thesis. Department ofAnthropology, Univ. of Vienna.

Gunz, P., and K. Harvati. 2007. The Neanderthal “chignon”: variation, inte-gration, and homology. J. Hum. Evol. 52:262–274.

Gunz, P., P. Mitteroecker, and F. Bookstein. 2005. Semilandmarks in threedimensions. Pp. 73–98 in D. E. Slice, ed. Modern morphometrics inphysical anthropology. Kluwer Press, New York.

Hallgrımsson, B., K. Willmore, and D. M. L. Cooper. 2004. Craniofacialvariability and modularity in macaques and mice. J. Exp. Zool. (Mol.Dev. Evol.) 302:207–225.

Hallgrımsson, B., D. E. Lieberman, W. Liu, A. F. Ford-Hutchinson, and F.

R. Jirik. 2007. Epigenetic interactions and the structure of phenotypicvariation in the cranium. Evol. Dev. 9:76–91.

Hanken, J., and J. B. Gross. 2005. Evolution of cranial development and therole of neural crest: insights from amphibians. J. Anat. 207:437–446.

Hansen, T. F. 2003. Is modularity necessary for evolvability? Remarks on therelationship between pleiotropy and evolvability. Biosystems 69:83–94.

Helms, J. A., D. Cordero, and M. D. Tapadia. 2005. New insights into cranio-facial morphogenesis. Development 132:851–861.

Herring, S. W. 1993. Epigenetic and functional influences on skull growth.Pp. 153–206 in J. Hanken and B. K. Hall, eds. The skull, Vol. 1. TheUniv. of Chicago Press, Chicago.

Hu, D., and J. A. Helms. 1999. The role of sonic hedgehog in normal andabnormal craniofacial morphogenesis. Development 126:4873–4884.

Jiang, X., S. Iseki, R. E. Maxson, H. M. Sucov, and G. M. Morriss-Kay. 2002.Tissue origins and interactions in the mammalian skull vault. Dev. Biol.241:106–116.

Khaitovich, P., I. Hellmann, W. Enard, K. Nowick, M. Leinweber, H. Franz, G.Weiss, M. Lachmann, and S. Paabo. 2005. Parallel patterns of evolutionin the genomes and transcriptomes of humans and chimpanzees. Science309:1850–1854.

King, M. C., and A. C. Wilson. 1975. Evolution at two levels in humansand chimpanzees. Their macromolecules are so alike that regulatorymutations may account for their biological differences. Science 188:107–116.

Klingenberg, C. P., K. Mebus, and J. C. Auffray. 2003. Developmental integra-tion in a complex morphological structure: how distinct are the modulesin the mouse mandible? Evol. Dev. 5:522–531.

Klingenberg, C. P., L. J. Leamy, and J. M. Cheverud. 2004. Integration andmodularity of the quantitative trait locus effects on geometric shape inthe mouse mandible. Genetics 166:1909–1921.

Kohn, L. A. P., S. R. Leigh, S. C. Jacobs, and J. M. Cheverud. 1993. Effects ofAnnular Cranial Vault Modification on the Cranial Base and Face. Am.J. Phys. Anthropol. 90:147–168.

Lande, R. 1979. Quantitative genetic analysis of multivariate evolution, appliedto brain:body size allometry. Evolution 33:402–416.

Lieberman, D. E., O. M. Pearson, and K. M. Mowbray. 2000a. Basicranialinfluence on overall cranial shape. J. Hum. Evol. 38:291–315.

Lieberman, D. E., C. Ross, and M. J. Ravosa.. 2000b. The primate cranial base:ontogeny, function, and integration. Yearb. Phys. Anthropol. 43:117–169.

Lieberman, D. E., B. M. McBratney, and G. E. Krovitz. 2002. The evolutionand development of cranial form in Homo sapiens. Proc. Natl. Acad. Sci.USA 99:1134–1139.

Manfreda, E., P. Mitteroecker, F. L. Bookstein, and K. Schaefer. 2006. Func-tional morphology of the first cervical vertebra in humans and nonhumanprimates. Anat. Rec. Part B 289:184–194.

Mann, A., and M. Weiss. 1996. Hominoid phylogeny and taxonomy: a consid-eration of the molecular and fossil evidence in a historical perspective.Mol. Phylogenet. Evol. 5:169–181.

Marie, P. J., J. D. Coffin, and M. M. Hurley. 2005. FGF and FGFR signaling inchondrodysplasias and craniosynostosis. J. Cell. Biochem. 96:888–896.

Marroig, G., and J. M. Cheverud. 2001. A comparison of phenotypic variationand covariation patterns and the role of phylogeny, ecology, and ontogenyduring cranial evolution of new world monkeys. Evolution 55:2576–2600.

Marroig, G., M. De Vivo, and J. Cheverud. 2004. Cranial evolution in sakis(Pithecia, Platyrrhini) II: evolutionary processes and morphological in-tegration. J. Evol. Biol. 17:144–155.

Mitteroecker, P. 2007. Evolutionary and developmental morphometrics of thehominoid cranium. Ph.D. Thesis. Department of Anthropology. Univ. ofVienna, Vienna.



Mitteroecker, P., and F. L. Bookstein. 2007. The conceptual and statisticalrelationship between modularity and morphological integration. Syst.Biol. 56:818–836.

Mitteroecker, P., P. Gunz, M. Bernhard, K. Schaefer, and F. Bookstein. 2004.Comparison of cranial ontogenetic trajectories among great apes andhumans. J. Hum. Evol. 46:679–697.

Mitteroecker, P., P. Gunz, and F. L. Bookstein. 2005. Heterochrony and geo-metric morphometrics: a comparison of cranial growth in Pan paniscusversus Pan troglodytes. Evol. Dev. 7:244–258.

Morriss-Kay, G. M. 2001. Derivation of the mammalian skull vault. J. Anat.199:143–151.

Moss, M. L., and R. W. Young. 1960. A functional approach to craniology.Am. J. Phys. Anthropol. 18:281–292.

Muller, G. B., and S. A. Newman. 2003. Origination of organismal form:beyond the gene in developmental and evolutionary biology. ViennaSeries in Theoretical Biology. MIT Press, Cambridge.

Needham, J. 1933. On the dissociability of the fundamental processes in on-togenesis. Biol. Rev. 8:180–223.

Nemeschkal, H. L. 1999. Morphometric correlation patterns of adult birds(Fringillidae: Passeriformes and Columbiformes) mirror the expressionof developmental control genes. Evolution 53:899–918.

Nie, X., K. Luukko, and P. Kettunen. 2006a. BMP signaling in craniofacialdevelopment. Int. J. Dev. Biol. 50:511–521.

———. 2006b. FGF signaling in craniofacial development and developmentaldisorders. Oral Dis. 12:102–111.

Noden, D. M. 1988. Interactions and fates of avian craniofacial mesenchyme.Development (Suppl.) 103:121–140.

O’Higgins, P. 2000. Quantitative approaches to the study of craniofacial growthand evolution: advances in morphometric techniques. Pp. 163–186 in P.O’Higgins and M. J. Cohn, eds. Development, Growth and Evolution.Academic Press, San Diego.

Ohno, S. 1970. Evolution by gene duplication. Springer, Berlin.Olson, E. C., and R. L. Miller. 1958. Morphological integration. Univ. of

Chicago Press, Chicago.Polanski, J. M., and R. G. Franciscus. 2006. Patterns of craniofacial integration

in extant Homo, Pan, and Gorilla. Am. J. Phys. Anthropol. 131:38–49.Radlanski, R. J., and H. Renz. 2006. Genes, forces, and forms: mechanical

aspects of prenatal craniofacial development. Dev. Dynam. 235:1219–1229.

Raff, R. 1996. The shape of life: genes, development, and the evolution ofanimal form. Univ. Chicago Press, Chicago.

Rhode, M. P., and B. T. Arriaza. 2006. Influence of cranial deformation onfacial morphology among prehistoric South Central Andean populations.Am. J. Phys. Anthropol. 130:462–470.

Riedl, R. J. 1978. Order in living organisms. John Wiley and Sons, New York.Riesenfeld, A. 1955. The variability of the temporal lines, its causes and effects.

Am. J. Phys. Anthropol. 13:599–620.Robinson, J. T. 1958. Cranial cresting patterns and their significance in the

Hominoidea. Am. J. Phys. Anthropol. 16:397–428.Rohlf, F. J., and M. Corti. 2000. The use of two-block partial least-squares to

study covariation in shape. Syst. Biol. 49:740–753.Rohlf, F. J., and D. E. Slice. 1990. Extensions of the procrustes method for the

optimal superimposition of landmarks. Syst. Zool. 39:40–59.Ross, C., and M. Henneberg. 1995. Basicranial flexion, relative brain size and

facial kyphosis in Homo sapiens and some fossil hominids. Am. J. Phys.Anthropol. 98:575–593.

Ross, C., and M. J. Ravosa. 1993. Basicranial flexion, relative brain size andfacial kyphosis in nonhuman primates. Am. J. Phys. Anthropol. 91:305–324.

Roth, V. L. 1996. Cranial integration in the sciuridae. Am. Zool. 36:14–23.Ruff, C. B. 1994. Morphological adaptation to climate in modern and fossil

hominids. Am. J. Phys. Anthropol. 37:65–107.Ruvolo, M. 1997. Molecular phylogeny of the hominoids: inferences from

multiple independent DNA sequence data sets. Mol. Biol. Evol. 14:248–265.

Santagati, F., and F. M. Rijli. 2003. Cranial neural crest and the building ofthe vertebrate head. Nat. Rev. Neurosci. 4:806–818.

Schlosser, G., and G. P. Wagner. 2004. Modularity in development and evolu-tion. The Univ. of Chicago Press, Chicago.

Schultz A. H. 1942. Condition for balancing the head in primates. Am. J. Phys.Anthropol. 29:483–497.

Schwartz, J. H., and I. Tattersall. 2002. The human fossil record, Vol. 1, ter-minology and craniodental morphology of the genus Homo (Europe).Wiley-Liss, New York.

———. 2003. The human fossil record, Vol. 2, craniodental morphology ofthe genus Homo (Africa and Asia). Wiley-Liss, New York.

Simon, H. A. 1962. The architecture of complexity. Proc. Am. Philos. Soc.106:467–482.

Sperber G. H. 2001. Craniofacial development. BC Decker Inc., Ontario.Strait, D. S. 2001. Integration, phylogeny, and the hominid cranial base. Am.

J. Phys. Anthropol. 114:273–297.Tapadia, M. D., D. Cordero, and J. A. Helms. 2005. It’s all in your head:

new insights into craniofacial development and deformation. J. Anat.207:461–477.

Terentjev, P. V. 1931. Biometrische Untersuchungen ber die morphologischenMerkmale von Rana ridibunda Pall. (Amphibia, Salientia). Biometrika23:23–51.

Trinkaus, E., and M. LeMay. 1982. Occipital bunning among later pleistocenehominids. Am. J. Phys. Anthropol. 57:27–35.

Von Dassow, G., and E. Munro. 1999. Modularity in animal development andevolution: elements of a conceptual framework for EvoDevo. J. Exp.Zool. (Mol. Dev. Evol.) 285:307–325.

Wagner, G. P. 1988. The influence of variation and developmental integrationon the rate of multivariate phenotypic evolution. J. Evol. Biol. 1:45–66.

———. 1990. A comparative study of morphological integration in Apis mel-lifera (Insecta, Hymenoptera). Z. zool. Syst. Evolut.-forsch. 28:48–61.

———. 1996. Homologues, natural kinds and the evolution of modularity.Am. Zool. 36:36–43.

Wagner, G. P., and L. Altenberg. 1996. Complex adaptations and the evolutionof evolvability. Evolution 50:967–976.

Wagner, G. P., J. G. Mezey, and R. Calabretta. 2005. Natural selection andthe origin of modules in W. Callebaut, ed. Modularity: understandingthe development and evolution of natural complex systems. MIT Press,Cambridge, MA.

Washburn, S. L. 1947. Relation of the temporal muscle to the form of the skull.Anat. Rec. 99:239–248.

Weidenreich, F. 1941. The brain and its role in the phylogenetic transformationof the human skull. Trans. Am. Phil. Soc. 31:321–442.

Wright, S. 1932. General, group and special size factors. Genetics 15:603–619.Zelditch, M. L. 1987. Evaluating models of developmental integration in the

laboratory rat using confirmatory factor analysis. Syst. Zool. 36:368–380.

———. 2001. Beyond heterochrony: the evolution of development. Wiley-Liss, New York.

Zelditch, M. L., and A. C. Carmichael. 1989. Ontogenetic variation in pat-terns of developmental and functional integration in skulls of Sigmodonfulviventer. Evolution 43:814–824.

Zhang, J. 2003. Evolution by gene duplication: an update. Trends Ecol. Evol.18:292–298.

Associate Editor: A. Badyaev



AppendixALGEBRAIC DETAILS

This appendix provides the algebraic details of the analysis of in-tegration and modularity. Let X = (XF |XN ) be the n " p matrix ofmean-centered shape coordinates, where XF is the n " p F matrixof shape coordinates for the face, XN the n " p N matrix for theneurocranium, with p = p F + p N , and n is the total sample size.Let further X# be the subset of adult specimens centered at theirspecies-specific group mean, that is, the ith specimen from the jthspecies is X#

i j = Xi j $ X j , where i ranges over all adult specimensand j = 1, 2, 3. The p F " p N matrix SF N = 1

n$1 (X#F )t X#

N , wherethe superscriptt denotes the matrix transpose, is thus the adultpooled within-species cross-block covariance matrix between thefacial and the neurocranial landmarks. Decompose this matrix asSFN = UDVt where D is a diagonal matrix of singular values. Thecolumns of U and V are the facial and neurocranial singular vec-tors, respectively, each representing one shape deformation thatis also called a SW (Bookstein et al. 2003). The scores along thefirst pair of singular vectors (the first pair of SW scores) have thehighest possible covariance, that is, Cov(X#

F U1, X#N V1) = d11 is

a maximum, where d11 is the first diagonal element of D, and U1

and V1 are the first columns of U and V. The scores along thesecond pair of vectors, constrained to be geometrically (but notstatistically) orthogonal to the corresponding first vectors, exhibitthe second highest covariance d22, etc. The maximal number ofSWs is min(p F , p N , n $ 1).

To decide how many dimensions of integration are worth an-alyzing, we tested the singular values dii against a permutationdistribution. Our method agrees with the procedure of Rohlf andCorti (2000) for the first test but differs for the subsequent di-mensions. To test the ith singular value, where i > 1, compute theresidual data matrices Z(i)

F =X#F $X#

F U1...(i$1) Ut1...(i$1) and Z(i)

N =X#

N $X#N V1...(i$1) Vt

1...(i$1), which are projections of the data ontothe subspaces perpendicular to the first i $ 1 singular vectors.The jth dimension of PLS between these two residual matricesis identical to the (i + j $ 1)th dimension of the original data.Permute the rows of one of the residual matrices and compute thefirst singular value for each permutation. By comparing the ithoriginal singular value with this distribution, we arrive at the sta-tistical significance level (the tail probability) of the ith dimensionof integration (the ith common factor).

In Mitteroecker and Bookstein (2007) we showed that theelements of the first pair of singular vectors are proportional to thecommon factor loadings in Sewall Wright’s (1932) factor analysiswhen both are based on the same assumptions about modularity,that is, on the same blocks of variables (except that for two modulesWright’s approach requires an additional scaling step as well).Thus, when scaled correctly, the pairs of corresponding singularvectors serve as common factors in the model outlined in Figure 1

(for a proof see Mitteroecker 2007). Loadings of the first principalcomponent of the blockwise common factor scores can be usedas the necessary scaling factors. Let C(i) be the n " 2 matrix ofscores (X#

F Ui |X#N Vi ) and e(i)

1 = (e(i)11, e(i)

12)t the first eigenvectorof (C(i))t C(i). The stacked vector,

fi =

!

"e(i)

11Ui

e(i)12Vi

#

$ ,

is then the ith common factor of X, with Ft F = I when F is takenas the p " k column vector matrix of k common factors (f1| . . .

|fk).The space of integrated shape variability (Fig. 3) is com-

puted as XF, where the number of common factors k was 4 inthe present analysis. The residual space R = X $ XFFt is thesubspace perpendicular to F, and the modular spaces for the faceand the neurocranium (Figs. 6 and 7) are principal componentscores of RF and RN , respectively. Figure 8 consists of plots ofXF Ui versus XN Vi for the ith dimension. Thus, even though PLSis based on the within-species variability only, the scores in thesefour figures are projections of the original data space and hencealso include between-species shape differences.

The values in the ith row of Table 1 are Cov(X#F Ui , X#

N Vi ),Cor(X#

F Ui , X#N Vi ), Var(X#Fi )/Tr(X#)" 100%, Var(XFi )/Tr(X) "

100%, and Cov(X#F Ui , X#

N Vi )2/% %

(S2F N )i j " 100%. The frac-

tion of modular variance along the ith common factor is computedas Var(RF Ui )/Var(XF Ui ) for the face and similarly for the neu-rocranium.

ALTERNATIVE ANALYSES

We carried out several alternative versions of these analyses tocheck the extent to which the results depend on certain assump-tions. The analyses were performed using the first 5, 10, and 20principal components, respectively, instead of the full Procrustescoordinates, and we varied the number of common factors from 2to 15. To enable the visualization of common factors as one jointshape deformation, the landmarks were separated in the presentanalysis into a facial and a neurocranial subset after an overall Pro-crustes fit. But we alternatively carried out separate Procrustes fitsfor the two modules. All these analyses yielded identical or verysimilar results.

As set out in the Methods section, the PLS analysis is basedon the pooled adult within-species covariance matrix. Alterna-tively, we separately performed analyses using the pooled adultwithin-sex, within-species covariance matrix (the matrix ignoringvariance and covariance among species means and among sexmeans); the adult chimpanzee covariance matrix; the adult humancovariance matrix; and the pooled adult chimpanzee within-sexcovariance matrix. Integrated and modular spaces remain verysimilar, but the actual common factors rotate when estimated



from different covariance matrices. However, similar conclusionsabout evolutionary integration and dissociation emerge from theseanalyses.

The present article focuses on the relationship between theface and the neurocranium but we also carried out the analysisassuming four modules (vault, cranial base, upper face, lowerface) instead of two (see Mitteroecker 2007 for an extension of

PLS to four blocks). Some dissociation was found within the face,especially between humans and nonhumans, whereas the cranialvault and base are quite integrated. This is reflected by the factthat common factors 1 and 3 of this article affect only the upperpart of the face whereas factor 4 involves only the lower part of theface. This local dissociation does not confound the conclusionsdrawn from the presented analysis.


THE EVOLUTIONARY ROLE OF MODULARITY AND INTEGRATION … · THE EVOLUTIONARY ROLE OF MODULARITY AND INTEGRATION IN THE HOMINOID CRANIUM Philipp Mitteroecker1,2,3,4 and Fred Bookstein1,3,5

Documents