The pattern of endocranial ontogenetic shape changes in humans

Abstract

Humans show a unique pattern of brain growth that differentiates us from all other primates. In this study, we use virtual endocasts to provide a detailed description of shape changes during human postnatal ontogeny with geometric morphometric methods. Using CT scans of 108 dried human crania ranging in age from newborns to adults and several hundred landmarks and semi-landmarks, we find that the endocranial ontogenetic trajectory is curvilinear with two bends, separating three distinct phases of shape change. We test to what extent endocranial shape change is driven by size increase and whether the curved ontogenetic trajectory can be explained by a simple model of modular development of the endocranial base and the endocranial vault. The hypothesis that endocranial shape change is driven exclusively by brain growth is not supported; we find changes in endocranial shape after adult size has been attained and that the transition from high rates to low rates of size increase does not correspond to one of the shape trajectory bends. The ontogenetic trajectory of the endocranial vault analyzed separately is nearly linear; the trajectory of the endocranial base, in contrast, is curved. The endocranial vault therefore acts as one developmental module during human postnatal ontogeny. Our data suggest that the cranial base comprises several submodules that follow their own temporally and/or spatially disjunct growth trajectories.

Introduction

One of the most important characteristics defining the genus Homo, and especially our own species Homo sapiens, is a large, complex and specialized brain. Associated with the unique adult form of the brain and its encasing bony structure is a unique pattern of growth that differentiates humans from all other primates (e.g. Bogin, 1999 in general; and Leigh, 2004 for the brain in particular). In this paper, we investigate human endocranial size and shape changes during postnatal ontogeny. Our aims are first to provide a detailed description of ontogenetic endocranial shape changes, and secondly to test hypotheses about (1) the relationship of size and shape changes during ontogeny and (2) about the pattern of endocranial form changes and its relationship to integration and modularity. Understanding human endocranial ontogeny will ultimately provide a framework for reliably interpreting brain evolution in fossil hominins.

Because brains themselves are not preserved in the fossil record, endocranial bones and endocasts (casts of the internal table of the neurocranium) can be used as a proxy of brain surface morphology in a comparative analysis of extinct hominins and living primates. To this end, we introduce a new methodology based on computer-generated endocasts and the tools of geometric morphometrics (Bookstein, 1991; Slice, 2007) that is also applicable to fossil material in future studies.

Background: brain – endocranium – endocast

Brain growth and development are linked tightly to the morphogenesis of the surrounding endocranium, especially in pre- and perinatal periods. The embryonic brain is enclosed by mesenchyme composed of two layers: the endomeninx and the ectomeninx (Sperber, 1989); the endomeninx forms the pia and arachnoid mater (Sperber, 1989). While the internal part of ectomeninx also remains unossified and forms the dura mater, the external osteo- and chondrogenic part contributes to bone formation (Moss & Young, 1960; Sperber, 1989). Therefore, the size and shape of endocranial bones also directly reflect the form of the adult brain. Analyses of phenotypic integration of the neurocranium and the brain in craniofacial pathologies support the notion that brain, meninges, and skull interact in highly coordinated and integrated ways (e.g. anencephaly, microcephaly, and craniosynostosis; Chervenak et al. 1984; Dambska et al. 2003; Richtsmeier et al. 2006).

When the brain expands, it generates tension along the endocranial surface of the braincase, especially via the falx cerebri and the tentorium cerebelli, which activates osteoblast deposition within sutures, and drift and endochondral growth in synchondroses (Moss & Young, 1960; Moss, 1962; Duterloo & Enlow, 1970; Lieberman, 1996; Lieberman et al. 2000a).

Because of the described interrelationships, endocranial bones and endocasts are often used as a proxy of brain surface morphology in studies of fossil material (‘paleoneurology’, see Holloway, 1978; Falk, 1980, 1986, 1987; Bruner, 2004; Holloway et al. 2004). Many of these comparisons are based on natural endocasts and artificially generated endocranial casts. More recently, so-called virtual endocasts generated from computer tomography (CT) scans of crania, have been studied (e.g. Conroy et al. 1990, 1998, 2000a,b; Falk et al. 2000, 2007; Bräuer et al. 2004; Neubauer et al. 2004; Falk & Clarke, 2007).

Exploratory analysis

Based on virtual endocasts, our first aim is to describe the endocranial shape changes during postnatal ontogeny from newborns to 12-year-olds and how shape variation within this ontogenetic sample relates to shape variation in adults. We use geometric morphometrics (GM) (Bookstein, 1991; Slice, 2007) based on Cartesian coordinates of anatomical landmarks as well as points on curves and surfaces (sliding semi-landmarks).

Relationship between size increase and shape changes

We also investigate the interrelationship of size increase and shape changes during human postnatal ontogeny. As the brain is growing, the braincase has to grow and develop accordingly. As outlined above, developmental shape changes are ‘constrained’ or ‘directed’ via dural tensors (falx cerebri, tentorium cerebelli) as well as basicranial parts that mature first and are integrated with the face (Biegert, 1963; Bastir et al. 2006).

Endocranial size changes

It is known that, prenatally, absolute brain size increases dramatically with age. Human neonatal brain size has been estimated at between 22 and 29% of adult size (DeSilva & Lesnik, 2006; Hublin & Coqueugniot, 2006). Postnatally, humans grow their brain rapidly at fetal growth rates within the first year or even longer (Leigh, 2004). Thereafter, growth rates decelerate and adult brain size is nearly (90–95%) achieved around 6 years of age; in terms of dental development this coincides with the time around M1 eruption (Holt et al. 1975; Gould, 1977; Martin, 1983; Smith & Tompkins, 1995; Coqueugniot et al. 2004; Leigh, 2004; Hublin & Coqueugniot, 2006). In comparison, neonates of other primates have achieved a larger percentage of adult brain size (e.g. about 40 ± 6% for chimpanzees, DeSilva & Lesnik, 2006) and grow their brain at fetal growth rates only for a short postnatal period (Coqueugniot et al. 2004; Gould, 1977; Holt et al. 1975; Hublin & Coqueugniot, 2006; Martin, 1983; Smith & Tompkins, 1995). However, the importance of this difference in postnatal brain growth at high rates as well as the duration of (subsequent slower) brain growth is a controversial subject (see Hublin & Coqueugniot, 2006; Jolicoeur et al. 1988; Leigh, 2004; Rice, 2002; Robson & Wood, 2008; Vrba, 1998).

Endocranial shape changes

As the brain is growing, its shape and the shape of the surrounding endocranium change. Of these shape changes, the cranial base morphology has been analyzed with traditional distance and angular measurements (e.g. Goodrich, 2005; Sgouros et al. 1999a,b); the midline aspect of ontogenetic and evolutionary changes in the cranial base has received particular attention. The uniquely human condition of an extremely flexed cranial base has been linked to various unique modern human traits such as a large brain relative to cranial base length (Weidenreich, 1941; Biegert, 1963; Enlow, 1968; Gould, 1977; Dean, 1988; Ross & Ravosa, 1993; Ross & Henneberg, 1995; Spoor, 1997), bipedal locomotion (Dabelow, 1929, 1931; Weidenreich, 1941; Schultz, 1960), facial retraction and orientation (Dabelow, 1929; Scott, 1958; Biegert, 1963; Enlow, 1976, 1990), and dimensions of the vocal tract and thereby spoken language (Lieberman et al. 1972; Laitman & Crelin, 1976; Laitman & Heimbuch, 1982; Lieberman et al. 1992).

According to one of these hypotheses, the ‘spatial packing hypothesis’, encephalization (brain size relative to cranial base length) causes a spatial packing problem that is solved by flexion of the cranial base between the pre- and postchordal planes and by coronal orientation of the petrous bones (e.g. Dabelow, 1931; Biegert, 1963; Gould, 1977; Ross & Ravosa, 1993; Lieberman et al. 2000b, 2008). This hypothesis is supported by the fact that basicranial flexion among non-human primates is highly correlated with brain size relative to cranial base length (Biegert, 1963; Gould, 1977; Ross & Ravosa, 1993). However, several fossil hominin specimens (e.g. Sts 5, OH 5, KNM-ER 3733, KNM-ER 3883, Bodo, Kabwe, Gibraltar 1, Guattari 1, Sambungmacan 4) exhibit cranial base angulation in the modern human range even though they show a wide range of cranial capacities (Ross & Henneberg, 1995; Spoor, 1997; Baba et al. 2003; Bastir & Rosas, 2009).

In an ontogenetic perspective, cranial base angulation in humans changes in a complex way while brain size increases rapidly. Prenatally, the embryonic cranial base flexes (Muller & O’Rahilly, 1980; Diewert, 1983; Sperber, 1989) but thereafter, during the fetal period, it retroflexes and at the same time the os petrosum orients coronally (Kvinnsland, 1971; Dimitriadis et al. 1995; Jeffery & Spoor, 2002). Jeffery (2002) showed that in human fetuses the supratentorial brain (the cerebrum located superior to the tentorium cerebelli) increases in size relatively more than the infratentorial brain and that this is correlated with inferoposterior rotation of the tentorium cerebelli. Postnatally, the cranial base flexes again until about 2 years of age and thereafter the angle is relatively stable (Lieberman & McCarthy, 1999), while brain size further increases. However, the midline cranial part does not only show angular modifications; its pre-sellar part becomes relatively elevated and its post-sellar part elongated inferiorly (Bastir et al. 2006). Moreover, the lateral cranial floor shows relative retraction of the anterior cranial fossa, relative forward projection as well as posterior expansion of the middle cranial fossa (Bastir et al. 2006). The shape of different endocranial parts matures far later than the midline angle: the midline cranial base at about 7–8 years, the midline neurocranium at about 9–10 years, and the lateral cranial floor at about 11–12 years, as measured from lateral X-rays of longitudinal data (Bastir et al. 2006).

Hypothesis

Because of these non-angular and off-midline shape changes, we investigate the interrelationship between size increase and shape changes in a more general way than the classic spatial packing hypothesis. We test to what extent overall postnatal endocranial shape change is driven by endocranial size increase. The null-hypothesis that endocranial shape change is exclusively driven by size increase, would be supported if shape matured together with size, and rejected if shape changed considerably after endocranial size is (almost) achieved at about 6 years of age or M1 eruption.

The pattern of endocranial form changes

We investigate the pattern of endocranial form changes throughout human postnatal ontogeny that we find in the exploratory analysis. In recent years, craniofacial shape changes during ontogeny have been analyzed as species-specific ontogenetic trajectories using multivariate methods. The discussion about cranial ontogenetic trajectories has evolved rapidly over the last decade. Initially, the discussion revolved around the question whether hominoid ontogenetic trajectories are parallel after birth (Richtsmeier & Lele, 1993; Richtsmeier & Walker, 1993; Bruner & Manzi, 2001; Ponce de León & Zollikofer, 2001; Ackermann & Krovitz, 2002; Penin et al. 2002; Zollikofer & Ponce de León, 2004) or not (Viðarsdóttir et al. 2002; Bastir & Rosas, 2004a,b; Cobb & O’Higgins, 2004, 2007; Mitteroecker et al. 2004a,b; Viðarsdóttir & Cobb, 2004). The later studies have shown that ontogenetic trajectories diverge and that therefore postnatal morphogenesis considerably contributes to adult craniofacial variation, accentuating the interspecific differences that are already established at birth. Most recently, several studies (Mitteroecker et al. 2004a,b; 2005; Bulygina et al. 2006; O’Higgins et al. 2006; Bastir & Rosas, 2009) have shown that these trajectories not only diverge, but that they are also not linear. Based on previous work (see references above, but most explicitly expressed in Bastir & Rosas, 2009), we expect the ontogenetic trajectory of the endocranium to be curvilinear as well.

Curvilinearity of ontogenetic trajectories can be interpreted with regard to morphological integration (Olsen & Miller, 1958) and developmental modularity. Analyses of intra- and interspecific craniofacial variation (e.g.Lieberman et al. 2000a; Lieberman et al. 2002; Bookstein et al. 2003; Bastir & Rosas, 2006, 2009; Gunz & Harvati, 2007; Bruner & Ripani, 2008) and of craniofacial pathologies and artificial skull deformations (e.g. Anton, 1989; Richtsmeier et al. 2006) show that the cranial base, the neurocranial vault, and the face act as tightly integrated modules. A module is a morphological unit that is characterized by more and stronger interactions within this unit than with other such units (Klingenberg, 2008 and references therein). This definition implies that ontogenetic changes in a module are highly integrated and therefore that the ontogenetic trajectory of one developmental module is expected to be fairly linear. In reverse, if morphological regions consisting of more than one module are analyzed together, the ontogenetic trajectory is expected to be curvilinear because different modules can have different spatial and temporal morphological modifications, i.e. different ontogenetic trajectories (cf. Bastir & Rosas, 2009).

Modular development of cranial base and vault

Here we test the most basic and straightforward model using two modules that have different embryological origins, the cranial base and the vault. The basicranium develops from endochondral ossification with growth in the spheno-ethmoid, midsphenoid, and spheno-occipital synchondroses, while most of the neurocranium grows from intramembranous ossification with deposition in sutures, and depository and resorptive drift (Scott, 1958; Duterloo & Enlow, 1970; Friede, 1981; Lieberman et al. 2000b). The basicranium is morphologically and developmentally conserved in mammals (de Beer, 1937; Bosma, 1976; Sperber, 1989). It is the first region to reach adult size and provides the platform on which the rest of the skull and the brain grow (Biegert, 1963).

Base and vault differ not only in their embryological origins but are also subject to different functional and spatial constraints. The cranial base, defined as the floor of the three cranial fossae, builds the platform on which the brain grows, contains foramina for important cranial nerves and vessels, and interacts with the spatially adjacent face, pharynx, and vertebral column. In contrast, the endocranial vault is less constrained and is able to grow in anterior, posterior, lateral, and superior directions to accommodate the growing brain.

Hypothesis

We hypothesize that the curvilinearity of the overall ontogenetic trajectory is caused by differences in growth rate and timing of endocranial vault and base. If the modules are identified correctly, their individual ontogenetic trajectories are linear when analyzed separately (cf. Mitteroecker et al. 2004b, 2005; Bastir & Rosas, 2009). This hypothesis would therefore be supported if the trajectories of vault and base analyzed separately were linear.

Methodological considerations

GM methods have several advantages over more traditional approaches using linear distances and angles. First, the geometric relationships among landmarks are preserved throughout the analysis and thereby results can be visualized intuitively. Secondly, it is possible to separate size from shape information and analyze these variables separately. Furthermore, so-called sliding semi-landmarks (Bookstein, 1997; Gunz, 2005; Gunz et al. 2005) make it possible to include curve and surface data from regions that lack traditional landmarks.

The GM approach for endocranial data used in this paper builds on previous work by Bookstein and colleagues, Bastir and colleagues and Bruner and colleagues (Bookstein et al. 1999, 2003; Bruner et al. 2003; Bruner, 2004; Bastir & Rosas, 2005; Bruner & Manzi, 2005, 2008; Bastir et al. 2006, 2008). As discussed in Bruner (2004), landmarks, which are defined on convolutions of the brain that left impressions into the endocranial bones, are hard to locate and occupy biologically homologous areas rather than precise point locations. Here, we use landmarks defined on endocranial bony structures and sliding semi-landmarks on curves and on surfaces. Surface semi-landmarks (Gunz et al. 2005) are particularly suited for the study of endocasts, as their surfaces are very smooth. In fact, Bruner's (2004) idea of measuring biologically homologous areas rather than precise point locations is the very definition of a surface semi-landmark. A surface semi-landmark ignores the information along the surface while retaining the geometric information perpendicular to the surface. For a discussion of methodological details, limitations, and alternative approaches of sliding see Gunz et al. (2005) and Perez et al. (2006). By not using landmarks defined on brain convolutions, our approach is limited in that it does not allow for exact determination of brain lobe boundaries. Surface semi-landmarks do, however, capture information of the lobe morphology itself. Therefore, shape changes and differences can be reported as changes in gross brain areas.

Materials and methods

Sample

The modern human sample used in this study comprises dried crania of a cross-sectional ontogenetic series (48 specimens) and of 60 adult specimens (Table 1). Immature specimens are in the age range between newborns and 12 years and come from two restricted geographical areas in Europe: a region at the border between France and Germany (Strasbourg anatomical collection) and a region comprising eastern Austria and Bohemia (Vienna anatomical collection). Calendar age and sex are taken from dissection reports or inscriptions directly on the crania. The adult sample (osteological collections of the University of Vienna, University of Leipzig, and University of Freiburg) includes specimens from Europe (n = 24) and craniometrically diverse specimens from the rest of the world (Africa, Asia, Australia, America, n = 36).

Table 1.

Human sample used in this study

Age group	Dentition	n (f/m)	Mean age (years)	Age range (years)
N	no teeth	9 (2/7)	0.13	0–0.33
NJ1	incomplete deciduous dentition	9 (5/4)	1.19	0.5–2
J1	complete deciduous dentition	21 (8/13)	3.55	1.83–5.33
J2	M1 erupted	9 (1/8)	9	6–12
A	M3 erupted	60 (27/33)
	thereof Europeans	24 (12/12)

Because dental eruption is strongly correlated with important life history variables (Smith, 1989), developmental age groups were established using dental eruption stages of the maxillary dentition. Specimens before the eruption of the first tooth are grouped into age group N. Age groups NJ1 and J1 comprise specimens with incomplete and complete deciduous dentition, respectively. Specimens that have an erupted M1 are assigned to group J2. After the M3 teeth have erupted, specimens are regarded as adults (age group A). Note that our sample does not contain age group J3 (M2 erupted).

Measurement protocol

CT data

All specimens were CT scanned with a pixel size between 0.24 and 0.49 mm and a slice thickness between 0.25 and 1.0 mm. Specimens from Leipzig were scanned with the industrial CT scanner system BIR ACTIS 225/300; all other specimens with medical CT scanners were available near their respective repositories (Siemens Sensation 16 and Siemens Plus 4 Volume Zoom). Measurements were obtained from virtual three-dimensional representations produced from these CT scans of the original specimens.

Virtual endocasts

Virtual endocasts were generated by a combination of two- and three-dimensional semi-automated segmentation in amira (Mercury Computer Systems Inc.). First, the cranium was segmented by setting a gray value range for bone. The three-dimensional segmented bone area was then artificially expanded by adding a predefined number of voxel layers on the cranial surface, producing a cranium with thickened cranial bones but small foramina and sutures closed. Along with manual segmentation to seal the foramen magnum this artificial expansion of the cranial bones facilitated segmentation of the endocast. Secondly, this endocast was allowed to grow in three dimensions to the same amount as the bone altered in the first step, so that the surface of the virtual endocast corresponds to the endocranial table of bone. Finally, the virtual endocast was corrected manually on single CT images in regions where the surface of the virtual endocast did not match exactly the surface of endocranial bone.

Endocasts generated using this procedure were compared to those generated by threshold segmentation on each CT image of the complete image stack. Comparison of 10 virtual endocasts (five immature, five adults) revealed that the differences were subtle and restricted to regions like the foramen magnum or the cribriform plate which have to be segmented manually with either method. There were, however, no major differences in cranial capacity obtained by the two methods (difference always less than 5 cm³ or 0.3%).

Landmark data

We digitized 29 anatomical landmarks defined on endocranial bony structures like sutures, foramina and points of maximum curvature (see legend of Fig. 1) as well as 278 sliding semi-landmarks on curves and the cerebral and cerebellar surfaces. The sphenoid curve establishes a border between the anterior and the middle cranial fossa and the petrous curve delineates the middle from the posterior cranial fossa. The transverse sinus curve forms the boundary between posterior cranial fossa and the vault. Curves on the basioccipital clivus and the foramen magnum capture the shape of the foramen magnum as well as basicranial angulation. Surface semi-landmarks and the midsagittal curve capture information of the cerebral and cerebellar shape. For an illustration of the landmark set see Fig. 1.

Fig. 1.

Illustration of landmark set. Rendered surface of a CT scan and a semi-transparent virtual endocast with endocranial landmarks shown as large, blue spheres, semi-landmarks on curves as blue, small spheres, and surface semi-landmarks as light blue, small spheres. Midsagittal (unpaired) landmarks: anterior sphenoid spine, foramen caecum, endobregma, endolambda, internal occipital protuberance, opisthion, basion, endosphenobasion, dorsum sellae. Bilateral (paired) landmarks: anterior clinoid process, optic canal, superior orbital fissure, foramen rotundum, foramen ovale, petrous apex, internal acoustic meatus, maximum curvature point between transverse and petrous curve, jugular foramen, hypoglossic canal. Landmarks in italics were allowed to slide on curves.

Acquisition of landmarks and semi-landmarks was done in amira by one observer (S.N.). Curves were sampled as densely spaced points along the curves; a template for semi-landmarks on the endocranial surface was digitized on one specimen.

Data preparation

Fifty-eight equidistant semi-landmarks were sampled on cubic splines of the densely spaced points digitized along the curves. For surfaces, there exists no analog of equidistant points. To achieve the same point count of surface semi-landmarks in corresponding areas of the endocast, the template of 220 surface semi-landmarks measured on one individual was warped to every specimen according to anatomical landmarks and curve semi-landmarks using the thin-plate-spline algorithm and then lofted onto the surface of the virtual endocast (following Gunz et al. 2005). Semi-landmarks were then allowed to slide along tangents to the curves and tangent planes to the surface so as to minimize the bending energy of the thin-plate spline interpolation function between each specimen and the Procrustes consensus configuration. This was done in an iterative process of sliding and projecting back onto the cubic splines of curves and the endocast surface until convergence (Bookstein, 1997; Bookstein et al. 1999; Gunz et al. 2005). Missing landmarks were fully relaxed during sliding (Gunz, 2005; Gunz et al. 2005; Gunz & Harvati, 2007). Some landmarks (the maximum curvature point between transverse and petrous curve, endobregma, and endolambda) were treated as landmarks in the first steps of distributing equidistant points on curves and surfaces but later treated as sliding semi-landmarks. For a more detailed discussion of this issue see Gunz et al. (2005). Endobregma and endolambda were allowed to slide because we are interested in the general shape of the cerebrum and not the relative contribution of individual cranial bones to this overall shape. After relaxation, semi-landmarks correspond geometrically among specimens within the sample and are used in multivariate shape analyses together with the anatomical landmarks.

Brain asymmetry is an interesting and important feature of endocranial morphology, but in this study we focused on the overall ontogenetic shape changes and ignored the asymmetry signal. Therefore, we symmetrized our data by computing the mean of the original and reflected relabeled specimens (Mardia et al. 2000; Bookstein, 2005); this makes the visualizations of shape changes easier to read.

Landmarks and slid semi-landmarks were superimposed using Generalized Least Squares Procrustes analysis (Gower, 1975; Rohlf & Slice, 1990), which removes information about location and orientation from the raw coordinates and scales each specimen to unit centroid size. Centroid size was computed as the square root of the summed squared Euclidean distances from each landmark to the centroid of the specimen (Dryden & Mardia, 1998). The resulting Procrustes shape coordinates and centroid sizes were used for statistical analyses.

Intra-observer error was tested by an analysis of repeated measurements. One of us (S.N.) measured the landmark set on one immature and one adult specimen three times each. The largest Procrustes distance between repeated measurements of the same individuals was about 2.5 times smaller than the smallest Procrustes distance between different specimens. Therefore, intra-observer error does not affect specimen affinity. Measurement error for each of the anatomical landmarks on average was under ±0.9 mm.

Analyses

The relationship between size and age was analyzed in a bivariate plot of centroid size against calendar age. Centroid size and cranial capacity do not scale in the same way because cranial capacity is a cubic measure. We therefore also provide a plot of relative cubed centroid size vs. age, which is comparable to published data showing cranial capacity relative to the adult mean against age.

Patterns of shape variation are analyzed using principal component (PC) analysis to reduce the dimensionality of the high-dimensional shape-space (Bookstein, 1991; Rohlf, 1993). To assess the pattern of the ontogenetic trajectory, we plotted a projection of the first three components that explain a large amount of the total variation. We also performed our analyses in Procrustes form-space, an extension of shape-space using the data matrix of shape variables augmented by one column of the logarithm of centroid size. This method was originally introduced as ‘size-shape-space’ (Dryden & Mardia, 1998; Mitteroecker et al. 2004a), but was renamed by the morphometric community at the Vienna Morphofest 2006. ‘Procrustes form-space’ should not be confused with a space constructed from a set of centered and rotated but not scaled coordinates also called form-space (Rohlf, 1996).

We separated the landmark set into two modules (endocranial base and endocranial vault) according to functional aspects (see above) and investigated ontogenetic trajectories for each module separately. For these analyses the semi-landmarks of each module were slid separately.

For visualization of ontogenetic shape changes, we used a series of shape differences between subsequent age group means. Whether the mean shapes and mean sizes were statistically different was tested using permutation tests (Good, 2000) of age group differences in size (differences in mean centroid size) and shape (Procrustes distance between group mean configurations). We computed 10 000 permutations; they were considered statistically significant at α ≤ 0.05. Data preparation and analyses were done in mathematica (Wolfram Research) using routines written by PG and Philipp Mitteroecker (Gunz et al. 2005, and for the mathematica code see appendix in Gunz, 2005).

Results

Size changes

Endocranial centroid size increases dramatically in the first 2 postnatal years (Fig. 2). Thereafter, growth rates decelerate with increasing age and endocranial size increases only slightly after age group J1 (Fig. 2a). Male adult size is on average larger than female adult size (Fig. 2). Sexual dimorphism during ontogeny cannot be discussed due to our sex-biased sample towards male specimens. In our sample, the youngest specimens have brain volumes about 20% of the adult values, and 90–95% of adult cranial capacity is achieved at about 7–8 years of age (Fig. 2b; relative cranial capacity and relative cubed centroid size vs. age).

Fig. 2.

Endocranial growth curve. Absolute endocranial centroid size (CS) plotted versus calendar age (left, a) and cubed CS relative to the adult mean plotted versus calendar age (right, b). Female specimens are shown as circles, males as squares. Solid growth curve, Bezier spline curve of CS against age; dashed growth curve, Bezier spline curve of cranial capacity against age. Age groups are separated by dashed vertical lines and adult females and males are shown separately. Note that relative cranial capacity (dashed curve, b) is nearly identical to relative cubed CS (solid curve, b).

The endocranial mean centroid sizes of age groups N, NJ1 and J1 are significantly different from the adult mean (group A), whereas mean centroid size of age group J2 is not (Table 2). The same is true when using cranial capacity measured in cm³ instead of centroid size because endocranial centroid size correlates highly with cranial capacity (Pearson's correlation coefficient r > 0.98).

Table 2.

Permutation tests of mean differences between age groups. P-values from permutation tests for mean differences in shape (upper right part of table) and size (lower left part of table) between age groups. Mean differences considered statistically significant at α ≤ 0.05 indicated by italics

	N	NJ1	J1	J2	A
N		0.0044	0.0001	0.0003	0.0001
NJ1	0.0006		0.0002		0.0001
J1	0.0001	0.0001		0.0060	0.0001
J2	0.0002		0.0013		0.0001
A	0.0001	0.0001	0.0001	0.1877

Shape changes

The first three PCs of endocranial shape-space explain about 53% of total shape variation. Figure 3 shows a projection of the three-dimensional space of these three PCs. The endocranial ontogenetic trajectory is curvilinear with a bend within age groups N and NJ1. Age groups NJ1, J1, and J2 plot consecutively along PC 1 but show some overlap. Furthermore, there is a second bend of the trajectory in late ontogeny. Immature specimens and European adults (filled triangles in Fig. 3) do not overlap at all in this space. Non-European adults from all over the world (opened triangles, Fig. 3) overlap with the European adults.

Fig. 3.

Endocranial shape-space. Projection of the first three components of shape-space (53% of total shape variation). Each point represents one specimen. Age group N: squares, NJ1: down triangles, J1: circles, J2: diamonds, A: up triangles (European specimens: filled symbols, other origins: opened symbols), age group means shown as larger symbols with a solid line connecting the series of mean shapes of subsequent age groups. Convex hulls (gray, transparent) are drawn for each age group. Variation of all adults is shown as a dotted convex hull. A quadratic regression on centroid size is shown as a gray line (for immature specimens only).

PC 1 captures the angulation of the cranial base in terms of the position and orientation of the anterior and posterior cranial fossae to each other and the orientation of the middle cranial fossae either more inferiorly or anteriorly directed as well as associated modifications of vault globularity. PC 2 describes the relationships among endocranial width, length, and height: an anteroposteriorly short endocast is wider and higher than an anteroposteriorly elongated one. PC 3 reflects shape changes in which lower and longer endocasts have more laterally positioned temporal poles, broader anterior cranial fossae, more anteriorly positioned clivus, and lower interpetrosal angles. Because PCs are mere projections of the original data (for a detailed technical discussion see Mitteroecker et al. 2005), a more comprehensive method for describing shape changes throughout ontogeny is to visualize mean shape differences between subsequent age groups in full shape-space (Fig. 4). These shape differences are summarized in Table 3 and discussed below. Mean shapes of all non-adult age groups are significantly different from adults and mean shapes of subsequent age groups are also significantly different (Table 2).

Fig. 4.

Endocranial shape differenes between age groups. A – lateral view, B – inferior view. 1 – Mean shape of age groups N to A from left to right (a–e). Endocranial surface shown as triangulated surface of all landmarks and semi-landmarks. Vectors show shape differences to the next age group, exaggerated by a factor of 5. 2 – Opaque surface corresponds to the exaggerated shape differences to the next age group. Transparent surface corresponds to the mean shape of this age group (as opaque surface in 1). Vectors as in 1. For a description of theses shape changes please see Table 3 and discussion.

Table 3.

Description of shape differences between subsequent age groups. For visualization see Fig. 4

age groups N–NJ1	Differences in the mean shapes between age groups N and NJ1 (Fig. 4a) are the expansion of cerebellar, lateral parietal, and posterolateral temporal areas relative to the occipital, midsagittal parietal, anterolateral and inferior temporal, frontal, and orbital areas. The parietal area develops parietal bossing. The temporal lobes rotate medially and frontal poles develop, the interpetrosal angle increases, and the posterior part of the foramen magnum rotates inferiorly and the anterior part gets broader. The prechordal plane becomes relatively shorter and the postchordal plane rotates (mainly by anterior rotation of the ‘inferior clivus’ from basion to sphenobasion), thereby causing overall midline basicranial flexion.
age groups NJ1–J1	Differences in the mean shape between age groups NJ1 and J1 (Fig. 4b) are the expansion of the cerebellar, occipital, lateral temporal, and mid/parasagittal frontal areas relative to the parietal, inferior temporal, lateral frontal, and orbital areas. Cerebellar and occipital poles develop. The temporal lobes become broader. The foramen magnum becomes rounder in shape. The prechordal plane shortens relatively and the angle between the planum sphenoideum and the cribriform plate begins to retroflex. The postchordal plane becomes relatively longer and rotates (mainly by elongation of the ‘superior clivus’ from sphenobasion to dorsum sellae) contributing to overall midline basicranial flexion.
age groups J1–J2	Differences in the mean shapes between age groups J1 and J2 (Fig. 4c) are the expansion of the inferior cerebellar area, and mid/parasagittal frontal areas relative to occipital, parietal, and orbital areas. The brain stem area with the foramen magnum expands inferiorly and the temporal poles rotate anteromedially. The prechordal plane lengthens relatively and the cribriform plate rotates superiorly. The postchordal plane lengthens relatively and rotates by elongation of both, the superior and inferior clivus and by rotation of the inferior clivus. Contrary modifications in the pre- and postchordal planes result in no changes in overall midline basicranial angulation.
age groups J2–A	Differences in the mean shapes between age groups J2 and A (Fig. 4d) are the expansion of the occipital, lateral temporal, temporal pole, prefrontal and orbital areas relative to the cerebellar, parietal and frontal areas. A posterior projection of the occipital area develops and the temporal poles rotate laterally. The endocast gets flatter and wider. The interpetrosal angle decreases slightly. The foramen magnum gets rounder. There is a retroflexion within the prechordal plane by superior rotation of the cribriform plate and inferior rotation of the planum sphenoideum. The postchordal plane lengthens and rotates contributing to overall midline basicranial elongation.

Form changes

Figure 5 shows projections of the first three PCs in Procrustes form-space. Here the endocranial ontogenetic trajectory (Fig. 5a) is elongated along the first principal component as compared to shape-space: PC 1 in form-space is highly correlated with centroid size (Pearson's correlation coefficient r ≈ 0.9963). We find the largest increase of centroid size within age groups N and NJ1 (Fig. 2) corresponding also to their large range of variation along PC 1 (Fig. 5a). The trajectory is curved with a bend within age groups N and NJ1. European adult specimens (filled triangles, Fig. 5a) are separated from immature specimens along PC 2 and PC 3, including another bend of the trajectory. Age group J2 completely overlaps with adults along PC 1, indicating that there are no size differences, but is completely separated along PC 2/3, indicating that there are shape differences between these two groups. The craniometrically diverse specimens from other geographical areas (opened triangles, Fig. 5a) extend the variation of the European adults (filled triangles, Fig. 5a) but do not cause a different pattern for late ontogeny.

Fig. 5.

Procrustes form-spaces. Projections of the first three components of (a) endocranial form-space, (b) form-space of cranial base only, (c) form-space of endocranial vault only. Each point represents one specimen. Age group N: squares, NJ1: down triangles, J1: circles, J2: diamonds, A: up triangles (European specimens: filled symbols, other origins: opened symbols), age group means shown as larger symbols with a solid line connecting the series of mean forms of subsequent age groups. Convex hulls (gray, transparent) are drawn for each age group. Variation of all adults is shown as a dotted convex hull (in a). A quadratic regression on centroid size is shown as a gray line (for immature specimens only). For a clearer picture only European adults are shown in (b) and (c).

When base and vault are analyzed separately, the ontogenetic trajectory of the endocranial base (Fig. 5b) is also curvilinear, showing the same bends as the trajectory of the entire endocranium. Age groups J2 and A overlap slightly. The ontogenetic trajectory of the endocranial vault (Fig. 5c), however, is nearly linear when analyzed alone and age group J2 completely overlaps with adults (even age group J1 does so partly).

Discussion

Ontogenetic size increase

The brain grows at high growth rates until about 2 years of age (see Fig. 2). Early postnatal ontogeny is therefore the crucial period for increasing the size of the brain after obstetrical constraints are overcome and after maternal energetic investments decrease in amount and become different in nature (see Martin, 1983, 1996; Rosenberg & Trevathan, 2001; Rosenberg & Trevathan, 2002). In our sample, 90–95% of adult brain size is achieved around 7– 8 years of age (see Fig. 2b) and the mean of centroid size in age group J2 (M1 but not M2 erupted) is not statistically different from the adult mean. Therefore, it can be concluded that most of adult brain size is achieved at about this age or around/after M1 eruption, consistent with earlier studies on human brain growth (Holt et al. 1975; Gould, 1977; Martin, 1983; Smith & Tompkins, 1995; Coqueugniot et al. 2004; Leigh, 2004; Hublin & Coqueugniot, 2006).

Ontogenetic shape changes

Endocranial morphogenesis during postnatal ontogeny can be divided into at least three distinct phases of different shape changes as assessed by the ontogenetic trajectory in the first three PCs of shape-space (Fig. 3).

Perinatal phase

The first phase from birth to about 1 year is characterized by high growth rates and the development of the posterior cranial fossa (Fig. 4a). This might be associated with relative growth of the cerebellum. Furthermore, the parietal and posterior temporal regions are expanding relative to the occipital and temporal regions (Fig. 4a). The neurocranial vault develops parietal bossing (Fig. 4a) and thereby becomes more globular. Development of the basioccipital clivus, first by rotation of the part inferior to the spheno-occipital synchondrosis (‘inferior clivus’) and secondly by rotation of the part superior to the synchondrosis (‘superior clivus’), results in overall midline basicranial flexion. At the same time, the interpetrosal angle increases (Fig. 4a). Using angle measurements, Lieberman & McCarthy (1999) have shown that the human cranial base flexes postnatally until about 2 years of age. This is confirmed by our data. In accordance with the spatial packing hypothesis, basicranial flexion, the reorientation of the petrous bones, but also relative enlargement of parietal and cerebellar areas are related to the high growth rates of the brain in this phase. However, the midline cranial base extends earlier, in the fetal period, when the brain also grows at high rates (Jeffery & Spoor, 2002).

Childhood phase

The second phase comprises infants, children, and juveniles until about 9 years of age. Brain growth rates are still high until about 2 years of age and then decrease with age until adult size is achieved. In this period, the shape of the temporal lobes are modified by medio-lateral as well as anterio-posterior expansion and rotation of the temporal poles (Fig. 4b,c). Medial parts of the frontal areas expand relatively. The foramen magnum becomes round in shape and moves inferiorly (Fig. 4b,c). While Lieberman & McCarthy (1999) found adult values of cranial base angulation from 2-year-olds on, the superior basioccipital clivus expands and the inferior clivus rotates further after 2 years but the cribriform plate and the planum sphenoideum also reorient, causing an extension and then a retroflexion within the prechordal plane. This is in line with Bastir et al.'s (2006) previous geometric morphometric analysis of midline cranial base shape, which showed that this structure matures in shape at about 7–8 years, far later than the adult angulation is achieved. Therefore, these morphological changes are not well captured by traditional midline angle measurements. However, midline cranial base morphology changes even if not measureable via angles while brain size still increases and thereafter when adult brain size has been attained.

Adolescent phase

Although our sample does not include adolescent individuals, there are pronounced shape differences between juvenile and adult specimens that we interpret here as a third phase of shape change during ontogeny. It seems unlikely that these shape differences are caused by a population ‘mismatch’ between the immature and the adult sample because inclusion of adult, craniometrically diverse specimens from all over the world does not alter this observation (Figs 3 and 5a). Major shape differences are the development of pronounced occipital poles in adults, a flatter and wider neurocranial vault, and modifications in temporal, orbital, and prefrontal regions (Fig. 4d). Further expansion and rotation of the postchordal plane and further retroflexion within the prechordal plane even cause a slight extension of midline basicranial morphology. This extension is small (approximately 3°) and should not be overinterpreted given the large adult variation in cranial base angles. Even though there are internal modifications of the brain until young adulthood (mainly late myelination in some brain regions, Giedd et al. 1999; Sowell et al. 1999; Durston et al. 2001; Paus et al. 2001), we consider it unlikely that these internal changes could cause the pronounced shape change of the outer endocast surface we find here. Instead, we speculate that adolescent shape changes of the endocranium are related to the integration with the viscerocranium. As temporal, orbital, and prefrontal brain regions (anterior and middle cranial fossae) are spatially close to the face and pharynx, numerous studies have analyzed and hypothesized about the interrelationships of these anatomical parts (e.g. Biegert, 1957; Ross & Ravosa, 1993; Ross & Henneberg, 1995; Lieberman et al. 2000a,b). Bastir & Rosas (2006) found a significant correlation between lateral cranial base and the face and only a tendency of nonsignificant correlated patterns between midline base and the face. They suggested that the effective interface between neurocranium and face might be the lateral basicranium rather than the midline cranial base. The lateral cranial base matures later than the midline base (Bastir et al. 2006) and therefore develops in conjunction with the also later maturing face. This leads to increased morphological covariation and at the same time mutual influences between these structures. Furthermore, Bastir et al. (2004) found that the petrosal part of the middle cranial fossa and the posterior mandibular ramus behave like a vertically aligned, integrated petroso-mandibular unit. Integrated development together with adjacent structures like the face or the mandible including complex spatiotemporal modular patterns might therefore be important especially for the (unexpected) late ontogenetic shape changes in the endocranium during adolescence, when brain size does not increase anymore. Further research including facial landmarks is required to assess this relationship.

Spatial packing – relationship between size and shape changes

We found basicranial flexion in the period of high growth rates, confirming previous analyses using cranial base angle measurements (Lieberman & McCarthy, 1999). Basicranial angulation is more or less constant after 2 years (Lieberman & McCarthy, 1999) but brain size still increases. In our 3D geometric morphometric approach we can show the shape changes in the midline aspect that do not affect the angle as well as shape changes in the lateral basicranial and neurocranial regions (see also Bastir & Rosas, 2006; Bastir et al. 2006; Bastir & Rosas, 2009).

The null-hypothesis that endocranial shape change is only driven by endocranial size increase is rejected on the basis of our data. Mean shapes of age group J2 and adults are significantly different and none of the immature specimens up to 12 year-olds has achieved morphology in the range of adult variation. At the same time, mean size of age group J2 is not significantly different from adult mean size and some specimens of age group J1 even have brain sizes in the adult range. In other words, endocranial size matures earlier than endocranial shape. Size and shape changes are dissociated during endocranial morphogenesis, consistent with size-shape dissociation found for different endocranial regions (Bastir et al. 2006). The bends of the shape trajectory do not correspond to the transition from high to low growth rates (high to low size increase). This further supports the notion that size and shape changes are dissociated during postnatal ontogeny. In early ontogeny, it seems likely that the size increase of the growing brain contributes to endocranial shape change, including cranial base flexion and petrous reorientation as suggested by the spatial packing hypothesis. However, endocranial shape is influenced by other factors than brain size, particularly during adolescence when size is not increasing anymore.

Our results, together with previous results of interspecific and intergeneric as well as pre- and postnatal data (e.g. Ross & Henneberg, 1995; Spoor, 1997; Lieberman & McCarthy, 1999; Jeffery & Spoor, 2002; Jeffery, 2003; Lieberman et al. 2008; Bastir & Rosas, 2009), do not support the hypothesis that basicranial flexion is caused mainly by solving the spatial packing problem of the growing brain. Midline basicranial flexion is only one aspect of basicranial morphology within the highly integrated craniofacial complex and it is likely that it is caused by a mixture of different mechanisms, i.e. spatial brain packing, facial packing, bipedal locomotion, and vocal tract dimensions (Dabelow, 1931; Weidenreich, 1941; Biegert, 1957; Moss, 1958; Scott, 1958; Schultz, 1960; Biegert, 1963; Enlow, 1976; Laitman & Crelin, 1976; Gould, 1977; Laitman & Heimbuch, 1982; Dean, 1988, 1990; Ross & Ravosa, 1993; Ross & Henneberg, 1995). In accordance with recent GM studies (Bastir et al. 2004, 2006; Bruner, 2004; Bastir & Rosas, 2005, 2006, 2009; Bruner & Ripani, 2008), our study shows that non-angular as well as off-midline shape changes are potentially relevant for all these mechanisms and interrelationships. Future studies should further investigate and test in a broader sense the hypotheses generated by traditional analyses of midline cranial base angles.

The pattern of ontogenetic form changes and modularity

The ontogenetic trajectory (both in shape-space and in form-space, Figs 3 and 5a) is curvilinear. This finding is consistent with studies of primate craniofacial morphogenesis (Mitteroecker et al. 2004a,b, 2005; Bulygina et al. 2006; O’Higgins et al. 2006; Bastir & Rosas, 2009). Linear trajectories reported by others (Richtsmeier & Lele, 1993; Richtsmeier & Walker, 1993; Bruner & Manzi, 2001; Ponce de León & Zollikofer, 2001; Ackermann & Krovitz, 2002; Penin et al. 2002; Zollikofer & Ponce de León, 2004) might be linear because very young individuals were not included in the analysis or because the region investigated acts as a single module throughout the investigated ontogenetic period.

It is possible to identify three distinct phases of different shape changes that are separated by two bends during postnatal morphogenesis of the endocranium. The first bend of the trajectory takes place within the period of high growth rates (age groups N and NJ1) and indicates that there is a shift in the modular shape changes (the pronounced localized shape changes of the posterior cranial fossa and relative expansion of the parietal areas are followed by different shape changes). This shift does not correspond to the transition from high to low growth rates. Furthermore, shape is changing differently during childhood and during adolescence, causing the second bend of the trajectory (localized shape changes in the temporal, orbital, and prefrontal regions become pronounced during adolescence). This shift of shape changes occurs at a time when there is barely any brain size increase.

We hypothesized that the curvilinearity of the trajectory is caused by modular endocranial development with the cranial base and the vault acting as two separate modules. The ontogenetic trajectories of these two modules analyzed separately should therefore be linear and only the combined analysis should lead to a curved trajectory. The trajectory of the endocranial vault is nearly linear. The overlap of age groups J1, J2 and A (Fig. 5c) shows that adult vault form is achieved earlier (around the maturation of adult brain size) than adult cranial base form (Fig. 5b). This is in line with the notion that the vault is (1) less constrained to accommodate the growing brain and (2) that vault form changes are mainly due to brain size increase (see functional matrix hypothesis, Moss & Young, 1960). Therefore, the vault can be interpreted as a developmental module during postnatal ontogeny as assessed by its ontogenetic trajectory. The trajectory of the cranial base, however, is curvilinear (Fig. 5b). Therefore, it does not act as one module or has different submodules at different times in postnatal ontogeny. The curvilinearity of the overall endocranial ontogenetic trajectory seems to be caused to a large extent by the curved trajectory of the cranial base. Our hypothesis is therefore only partly supported, insofar as the vault acts as a developmental module but not that the curvilinearity of the overall endocranial trajectory is caused by modular development of the endocranial vault and base.

On a lower hierarchical level it can be similarly hypothesized that the curvilinear trajectory of the cranial base is caused by modular growth. Bastir & Rosas (2009) showed a curved ontogenetic trajectory for the endocranial base with at least four morphologically different phases during pre- and postnatal ontogeny. They described two postnatal phases, the first from birth until M1 eruption with integration of the greater sphenoid wings and the petrosals relative to the midline base, the second thereafter until adulthood with modular changes of the lateral basicranium relative to the midline (Bastir & Rosas, 2009). Our data point to a further subdivision of their first postnatal phase, as the ontogenetic trajectory in our study (Fig. 5b) shows a slight bend within this phase. Bruner & Ripani (2008) concluded in their analysis of morphological variation of the endocranial base that the three endocranial fossae are integrated only to a small extent. Large and local changes of the posterior fossa influence the rest of the endocranium (especially the position of the anterior cranial fossa) via the dural tensor, the falx cerebri (Bruner & Ripani, 2008). The cranial fossae seem to act as modules, which are then integrated at a higher hierarchical level. This interpretation matches our findings of a curved ontogenetic trajectory for the cranial base.

The curvilinearity points to complex patterns of covariation between the endocranial fossae and the midline and lateral aspects of the cranial base as found by other studies (Bruner & Ripani, 2008; Bastir & Rosas, 2009). However, the interrelationships within the cranial base and with other craniofacial structures are far from being resolved and exceed the scope of the present paper. Future studies should include partial least squares analyses to investigate the integrated and modular relationships of the cranial base and to provide functional and evolutionary hypotheses.

Evolutionary implications

In this study, we analyzed the endocranial ontogenetic pattern of humans and showed that our methodological approach is useful for research of endocranial morphology. Fossil hominin specimens can be included in this methodological framework. But because the record of well-preserved fossils is small, it is important to understand the endocranial ontogenetic patterns of living hominids and their differences before interpreting fossil specimens. Comparison of ontogenetic trajectories among living hominids can be used to inform analyses about the evolution of uniquely human features of the endocranium or brain. We found distinct phases of different shape changes for humans in this study. Therefore, we speculate that prolongation, truncation or shifts of one or more ontogenetic phases could have been important during hominin evolution.

Summary

Using CT scans to gain non-destructive access to the endocranium and sliding semi-landmarks in a GM analysis to quantify endocranial morphology adds shape as an important aspect to endocranial ontogenetic research. In this paper, we investigated the pattern of human endocranial morphogenesis during postnatal ontogeny and the interrelationship of size and shape changes.

We conclude that the human endocranial morphogenesis is characterized by a curvilinear trajectory that can be divided into distinct dissociated phases of size and shape change. The curvilinearity is a result of different localized form changes or, in other words, by modular development of different endocranial parts. While we identified the endocranial vault as one developmental module, our data corroborate the idea that the cranial base consists of several submodules (Bruner & Ripani, 2008; Bastir & Rosas, 2009). Spatial packing of the growing brain may contribute to endocranial shape changes in early ontogeny, but the hypothesis that endocranial shape change is exclusively driven by endocranial (and thus brain) size increase is not supported by our data. We find late endocranial shape changes in temporal, orbital, and prefrontal regions after endocranial size has been achieved. These changes might be caused by the integration with adjacent structures like the face.

The methodological framework established in this study will contribute to the understanding of hominin brain evolution when applied to the fossil record in future studies.