A new OH5 reconstruction with an assessment of its uncertainty Stefano Benazzi a, * , Fred L. Bookstein a, b , David S. Strait c , Gerhard W. Weber a a Department of Anthropology, University of Vienna, Althanstrasse 14, 1090 Vienna, Austria b Department of Statistics, University of Washington, Seattle, WA 98195, USA c Department of Anthropology, University at Albany,1400 Washington Ave, Albany, NY 12222, USA article info Article history: Received 15 March 2010 Accepted 7 February 2011 Keywords: Virtual anthropology Paranthropus boisei Geometric morphometrics Computer-aided design Fossil reconstruction abstract The OH5 cranium, holotype of Paranthropus boisei consists of two main portions that do not fit together: the extensively reconstructed face and a portion of the neurocranium. A physical reconstruction of the cranium was carried out by Tobias in 1967, who did not discuss problems related to deformation, although he noted a slight functional asymmetry. Nevertheless, the reconstructed cranium shows some anomalies, mainly due to the right skewed position of the upper calvariofacial fragment and uncertainty of the relative position of the neurocranium to the face, which hamper further quantitative analysis of OH5 0 s cranial geometry. Here, we present a complete virtual reconstruction of OH5, using three- dimensional (3D) digital data, geometric morphometric (GM) methods and computer-aided design (CAD) techniques. Starting from a CT scan of Tobias’s reconstruction, a semi-automatic segmentation method was used to remove Tobias’s plaster. The upper calvariofacial fragment was separated from the lower facial fragment and re-aligned using superposition of their independent midsagittal planes in a range of feasible positions. The missing parts of the right hemiface were reconstructed using non-uniform rational basis-spline (NURBS) surface and subsequently mirrored using the midsagittal plane to arrive at a symmetrical facial reconstruction. A symmetric neurocranium was obtained as the average of the original shape and its mirrored version. The alignment between the two symmetric shapes (face and neurocranium) used their independent midsagittal plane and a reference shape (KNM-ER 406) to highly reduce their degrees of freedom. From the series of alternative reconstructions, we selected the middle of this rather small feasible range. When reconstructed as a range in this way, the whole cranial form of this unique specimen can be further quantified by comparative coordinate-based methods such as GM or can be used for finite element modeling (FEM) explorations of hypotheses about the mechanics of early hominin feeding and diets. Ó 2011 Elsevier Ltd. All rights reserved. Introduction In July of 1959, Leakey (1959) found fragments of a new fossil cranium from Bed I of Olduvai Gorge, now dated to approximately 1.85 Ma (Tamrat et al., 1995; Hay and Kyser, 2001; Wood and Costantino, 2007). This subadult cranium, which became the holotype of Zinjanthropus boisei (now typically attributed to the genus Paranthropus), was first reconstructed and described by Louis and Mary Leakey (Leakey, 1959). Subsequently, Tobias (1967) slightly modified the reconstruction, while at the same time providing an exhaustive study of the fossil. The fossil consists of two main portions that do not have direct contact: the recon- structed face, consisting of a maxillofacial and a craniofacial fragment (Fig. 1a), and a large portion of the neurocranium comprising a biparietal fragment, the right temporal bone, the left temporal bone and an occipitosphenoid fragment (Fig. 1b). Both Leakey (1959) and Tobias (1967) stated that the fossil had not been affected by deformation, nor has any subsequent worker claimed postdepositional distortion of the cranium (e.g., Rak, 1983; Wood, 1991; Wolpoff, 1999; Schwartz and Tattersall, 2005). On the other hand, craniofacial asymmetry is generally recognized in the differential wear pattern of the right maxillary teeth compared with the left teeth and in the tilting of the midsagittal crest toward the right side. Both of these features could have arisen in life from the larger force generated by the right masseter and temporal muscles (Tobias, 1967; Rak, 1983). Tobias (1967) emphasized that the lack of distortion and the generally good condition of the fossil facilitated his reconstruc- tion of the two facial fragments as well as the neurocranial portion. The major task for him was to find the best estimate for * Corresponding author. E-mail address: [email protected](S. Benazzi). Contents lists available at ScienceDirect Journal of Human Evolution journal homepage: www.elsevier.com/locate/jhevol 0047-2484/$ e see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.jhevol.2011.02.005 Journal of Human Evolution 61 (2011) 75e88
A new OH5 reconstruction with an assessment of its uncertainty
Stefano Benazzi a,*, Fred L. Bookstein a,b, David S. Strait c, Gerhard W. Weber a
aDepartment of Anthropology, University of Vienna, Althanstrasse 14, 1090 Vienna, AustriabDepartment of Statistics, University of Washington, Seattle, WA 98195, USAcDepartment of Anthropology, University at Albany, 1400 Washington Ave, Albany, NY 12222, USA
a r t i c l e i n f o
Article history:Received 15 March 2010Accepted 7 February 2011
0047-2484/$ e see front matter � 2011 Elsevier Ltd.doi:10.1016/j.jhevol.2011.02.005
a b s t r a c t
The OH5 cranium, holotype of Paranthropus boisei consists of two main portions that do not fit together:the extensively reconstructed face and a portion of the neurocranium. A physical reconstruction of thecranium was carried out by Tobias in 1967, who did not discuss problems related to deformation,although he noted a slight functional asymmetry. Nevertheless, the reconstructed cranium shows someanomalies, mainly due to the right skewed position of the upper calvariofacial fragment and uncertaintyof the relative position of the neurocranium to the face, which hamper further quantitative analysis ofOH50s cranial geometry. Here, we present a complete virtual reconstruction of OH5, using three-dimensional (3D) digital data, geometric morphometric (GM) methods and computer-aided design (CAD)techniques. Starting from a CT scan of Tobias’s reconstruction, a semi-automatic segmentation methodwas used to remove Tobias’s plaster. The upper calvariofacial fragment was separated from the lowerfacial fragment and re-aligned using superposition of their independent midsagittal planes in a range offeasible positions. The missing parts of the right hemiface were reconstructed using non-uniformrational basis-spline (NURBS) surface and subsequently mirrored using the midsagittal plane to arrive ata symmetrical facial reconstruction. A symmetric neurocranium was obtained as the average of theoriginal shape and its mirrored version. The alignment between the two symmetric shapes (face andneurocranium) used their independent midsagittal plane and a reference shape (KNM-ER 406) to highlyreduce their degrees of freedom. From the series of alternative reconstructions, we selected the middle ofthis rather small feasible range. When reconstructed as a range in this way, the whole cranial form of thisunique specimen can be further quantified by comparative coordinate-based methods such as GM or canbe used for finite element modeling (FEM) explorations of hypotheses about the mechanics of earlyhominin feeding and diets.
� 2011 Elsevier Ltd. All rights reserved.
Introduction
In July of 1959, Leakey (1959) found fragments of a new fossilcranium from Bed I of Olduvai Gorge, now dated to approximately1.85 Ma (Tamrat et al., 1995; Hay and Kyser, 2001; Wood andCostantino, 2007). This subadult cranium, which became theholotype of Zinjanthropus boisei (now typically attributed to thegenus Paranthropus), was first reconstructed and described by Louisand Mary Leakey (Leakey, 1959). Subsequently, Tobias (1967)slightly modified the reconstruction, while at the same timeproviding an exhaustive study of the fossil. The fossil consists oftwo main portions that do not have direct contact: the recon-structed face, consisting of a maxillofacial and a craniofacial
nazzi).
All rights reserved.
fragment (Fig. 1a), and a large portion of the neurocraniumcomprising a biparietal fragment, the right temporal bone, the lefttemporal bone and an occipitosphenoid fragment (Fig. 1b).
Both Leakey (1959) and Tobias (1967) stated that the fossil hadnot been affected by deformation, nor has any subsequent workerclaimed postdepositional distortion of the cranium (e.g., Rak, 1983;Wood, 1991; Wolpoff, 1999; Schwartz and Tattersall, 2005). On theother hand, craniofacial asymmetry is generally recognized in thedifferential wear pattern of the right maxillary teeth comparedwith the left teeth and in the tilting of the midsagittal crest towardthe right side. Both of these features could have arisen in life fromthe larger force generated by the right masseter and temporalmuscles (Tobias, 1967; Rak, 1983).
Tobias (1967) emphasized that the lack of distortion and thegenerally good condition of the fossil facilitated his reconstruc-tion of the two facial fragments as well as the neurocranialportion. The major task for him was to find the best estimate for
Figure 1. The three main fragments of OH5 according to Tobias (1967), pls. 10, 15, 16: a) maxillofacial (bottom) and craniofacial (top) fragment (the arrow points to the infero-lateralmargin of the right orbit in the maxillofacial fragment); b) neurocranium.
S. Benazzi et al. / Journal of Human Evolution 61 (2011) 75e8876
the position of these three parts with respect to one another. Tothis end, he identified a contact between the two facial fragmentsin the infero-lateral part of the right orbital margin (Tobias, 1967).In addition, he paid attention to the “co-planarity of the inter-nasal suture above with the intermaxillary suture below, as wellas by the relationship between the temporal surfaces of thefrontal bone above, and those of the zygomatic and maxillarybones below” (Tobias, 1967, 8). The position of the left lateral
Figure 2. a) Tobias’s reconstruction of the OH5 face, approximately oriented according to tx-axis of the occlusal plane) highlights the dislocation of the lower margin of the left orbit c(1967), without plaster. The angle between the midsagittal plane of the maxillofacial fragmcontact (the frontal process of the right zygomatic) of the two fragments.
pterygoid lamina of the facial portion with the left residual part ofthe greater wing of the sphenoid bone (in the neurocranium)provided him with some indications regarding the relative posi-tion of the two main parts (the face and the neurocranium).Tobias stressed that “the temporal process of the right zygomaticbone is intact as far back as the zygomatic suture, while thezygomatic process of the left temporal bone is intact as farforward as the same suture” (Tobias, 1967, 8).
he occlusal plane of the maxillary teeth. The white line (approximately parallel to theompared with the right one. b) 3D digital model of OH5 face as reconstructed by Tobiasent (white) and craniofacial fragment (red) is 2.91�. Inset: enlargement of the region of
Figure 3. Phases of the analysis.
S. Benazzi et al. / Journal of Human Evolution 61 (2011) 75e88 77
As the best known individual of Paranthropus boisei, OH5 is oneof the most important specimens representing robust austral-opiths. Its thickly enameled molars are among the largest of anyhominin, and the skull shows adaptations to generate and with-stand large chewing forces to such an extent that in the popularliterature, OH5 is often referred to as the ‘Nutcracker Man.’ It hasplayed a central role in modeling masticatory biomechanics as wellas in understanding the evolution of the form of this branch ofaustralopiths (e.g., O’Higgins et al., 2010; Wroe et al., 2010).However, when looking at a frontal view (in Fig. 2a OH50s face ismanually oriented according to the occlusal plane of the maxillaryteeth), the upper face seems to be tilted a bit to the right, and theright and left orbits are clearly of different shape (cf. Neubauer et al.,2005). When all of the original pieces were put together duringphysical reconstruction, the upper face was obviously positioned
Figure 4. Seven facial versions obtained by moving the aligned upper fragment by 1 mm s
slightly askew.Moreover, there is uncertainty regarding the relativeposition of the neurocranium to the face; uncertainty that mustsignificantly affect estimates of cranial capacity. This does nothamper morphological description per se, nor is it meant as a crit-icism of Tobias (1967), whose description of the specimen wasa landmark contribution to paleoanthropology in his time.However, today we have more advanced methods available for theanalysis of morphology. For instance, quantitative analyses of formand shape, as carried out in Virtual Anthropology (VA), needaccurate coordinate-based data that preferably exclude deforma-tions that were not present in the living individual. Another tech-nique at our disposal in the 21st century is biomechanical modelingvia finite element analysis (FEA). Strains and stresses, for instancethose occurring on the skull during mastication, can be simulatedbymeans of solid models (e.g., Strait et al., 2009). Those simulations
teps in the z-direction. Splines are digitized by hand along the left profile of the nose.
Figure 5. a) Template of 104 (semi)landmarks on the mirrored left hemi-upper fragment. Black spheres indicate the locus of temporal line semilandmarks that will be estimated inthe target. b) Based on the estimated semilandmarks (black spheres) the upper face was 1.5 mm upward displaced for the sake of temporal line continuity. Arrow: contact areabetween the two fragments in the frontal process of the zygomatic bone. Black surface: reconstruction of right side as mirrored left side through the calculated midsagittal plane(see text). c) Curves digitized around the missing area. d) NURBS surface created from the curves network. e) The reconstructed surface as “merged” to the right hemiface.
S. Benazzi et al. / Journal of Human Evolution 61 (2011) 75e8878
need the entire surface geometry to be able to create solid models.Correction of deformation is crucial along with information on theuncertainty remaining in the reconstructed form. There is thusa definite purpose for such a reconstruction as we introduce it here,both for the quantitative comparison of form as well as for studyingdietary adaptations of robust australopiths.
Starting with the three pieces available to Tobias, we assign sixdegrees of freedom (three for translation, three for rotation) forpositioning the lower and upper face to each other, and sixothers formatching the face to the braincase. We take advantage of the exactcontrol of repositioning that is so straightforward in the virtualapproach to reconstruction (Weber and Bookstein, 2010). As shownin Fig. 2a, the lower margin of the left orbit (and hence the left
hemi-craniofacial fragment) is displaced upward with respect to theright homolog. As the fragments are not distorted, we attribute thisskewing to an incorrect position of the craniofacial fragment.
In this paper, we apply the toolkit of VA to reassemble the OH5cranium using state-of-the-art methods and demonstrate thecapabilities of such an approach with regard to documentation(making the process reproducible) and assessment of uncertainty.The potential of the virtual approach for fossil reconstruction hasbeen emphasized on various grounds (Kalvin et al., 1995; Zollikoferand Ponce De León, 2005; Ogihara et al., 2006; Gunz et al., 2009;Neeser et al., 2009). Among other published justifications, thefossil can be virtually manipulated without tampering with theoriginal, artificial material can be removed electronically, a range of
Figure 6. Reconstructed OH5 face mirroring the right side along the midsagittal plane: a) anterior view; b) posterior view.
S. Benazzi et al. / Journal of Human Evolution 61 (2011) 75e88 79
alternative solutions can be provided, and all phases are highlycontrollable and thus repeatable. Starting from the reconstructionof the three main fragments accomplished by Tobias (1967), weused 3D digital data, geometric morphometric (GM) methods, andcomputer-aided design (CAD) techniques to narrow the range ofplausible solutions for a symmetrized reconstruction of OH5. Wereport this range and select one preferred representative fromwithin it.
Materials and methods
Fig. 3 enumerates the phases involved in our reconstruction.
Figure 7. Neurocranium. a) right lateral view; b) left lateral view; c) the mirror of the right sibone (dark surface) and the mirror (light grey).
Phase 1
OH50s face and neurocranium were scanned using ComputedTomography at the Radiologie 2 Department at Medizinische Uni-versität Innsbruck, Austria (Siemens Somatom Plus 40). CT scanswere recorded in DICOM file format at a reconstructedmatrix size of512 � 512 pixels. Pixel size was 0.48 mm and slice thickness was1 mm. This virtual specimen can be purchased at http://www.virtual-anthropology.com/3d_data/3d-archive. 3D reconstructionsfor the face and the neurocranium were built using Amira 5.2 soft-ware (Mercury� Computer Systems, Chelmsford, MA). The modelswere achieved by manual segmentation, contour extraction, and
de creates a “double crest” on both side; d) step of 2.3 mm between the original parietal
Figure 8. a) Complete template with anatomical landmarks (black; n ¼ 11) and curve and surface semilandmarks (grey; n ¼ 186). Labels as in Table 1. b) Symmetric reconstructionof the neurocranium. The midsagittal plane and reconstructed zygomatic arches are in black.
S. Benazzi et al. / Journal of Human Evolution 61 (2011) 75e8880
surface reconstruction, removing the plaster and the matrix thatadhere to the original fossil.
Since the face and the neurocranium are not in direct contact,each part was symmetrized separately prior to any alignment. Wedescribe the symmetric reconstruction of the face, then thesymmetric reconstruction of the neurocranium, and finally thealignment process.
Table 1OH5 neurocranium: list of anatomical landmarks and curves of the template.
Anatomical landmarks N Curve names Smlmcounta
Auriculare posterior left (apl) 1 Median nucal line 5Auriculare posterior right (apr) 2 Midsagittal crest 9Basion (ba) 3 Superior nucal line left 10Condylus occipitalis posterior
left (copl)4 Superior nucal line right 10
Condylus occipitalis posteriorright (copr)
Total semilandmarks oncurves
34
Inion (i)Lambda (l)Medial-glenoid left (mgl)Medial-glenoid right (mgr)Vaginal process left (vpl)Vaginal process right (vpr)
a Semilandmarks identified on the curves.
Phase 2: OH5 faciocalvarial portion
Fig. 2b shows the 3D reconstruction of the fossil freed of plaster.In Tobias’s reconstruction, the infero-lateral margin of the rightorbit was included in the maxillofacial fragment (or lower frag-ment, Fig. 1a). We consider it more appropriate to attribute theorbital process of the right zygomatic bone to the craniofacialfragment (upper fragment). In fact, as mentioned by Tobias (1967)himself and further verified after segmentation, this bone makesgood contact with the upper fragment in the infero-lateral part ofthe right orbit, but is clearly separated from the lower fragment(Fig. 2b).
In the IMEdit module of PolyWorks� 10.1 (InnovMetric Soft-ware, Inc.), a midsagittal plane was defined both for the upper andthe lower fragment of the faciocalvarial portion. This was accom-plished by selecting points along the midsagittal region and thencomputing the best-fit plane (by standard least-squares algorithm)passing through these points. For the lower fragment, threeanatomical landmarks (prosthion, orale, staphylion) and an addi-tional 15 points digitized along the maxillary and palatine suture,respectively, were used (Fig. 2b, white line). For the upper frag-ment, three anatomical landmarks (rhinion, nasion, foramencecum) and 10 points were digitized along the nasal sutures andthe frontal crest (Fig. 2b, red line). The two midsagittal planes lie atan angle of 2.91 � to one another. We aligned the upper fragment tothe lower by superimposing the pair of midsagittal planes upon theyz-plane of our Cartesian coordinate system. After this superposi-tion, there are only 3� of freedom left: upwardedownward trans-lation (z-axis), forwardebackward translation (y-axis), andclockwiseecounterclockwise rotation about the x-axis. We rest-ricted the range of the latter two parameters so as to maintaincontact between the surfaces of the upper and lower fragment inthe right frontal process of the zygomatic bone, as they appear notto have been bent in opposite directions at any time post-mortem.The major uncertainty remaining involves the z-axis translation ofthe upper fragment.
We converge on a unique preferred value of this remainingdegree of freedom using either of two approaches. In the first
approach, the aligned upper fragment was translated along the z-axis by 1 mm steps to a range of �3 mm, resulting in seven facialalternatives (Fig. 4). In RapidForm XOR2 (INUS Technology, Inc.),splines were digitized by hand along the left profile of the nose. Themore the upper fragment is downwardly displaced, the more theprofile of the nose bends mesially, leading to a shape that seems nolonger to display a smooth curvature. If one accepts this assumptionof a smooth curvature of the nasal aperture, an upward displace-ment of the upper fragment is deemed more consistent witha natural shape than a downward displacement.
In the second approach, we used information from the fronto-temporal region of the left side to reconstruct by thin-plate spline(TPS) the portion missing on the right side. In the IMEdit module ofPolyWorks, the left hemi-upper fragment was mirrored along itsmidsagittal plane (used here as a reference model, see Weber andBookstein, 2010) and three anatomical landmarks (rhinion (rh),foramen cecum (fc), frontomalare-orbitale (fmo)) and 101 semi-landmarks on curves (n ¼ 37) and surfaces (n ¼ 64) were located touse as a template (Fig. 5a). The three curves were digitized alongthe upper (one) and the lower (two) margins of the orbit (theorbital curves started from frontomalare-orbitale), as well as on thefrontotemporal line (three) (Fig. 5a). Using the open-source soft-ware Edgewarp3D (Bookstein and Green, 2002), the template waswarped onto the right hemi-upper fragment (target) by iterativeTPS. The template is aligned to the target according to theanatomical landmarks. Semilandmarks were allowed to slide alongcurves and surfaces to minimize the bending energy of the TPSinterpolation function computed between the reference and thetarget (Bookstein, 1991; Gunz et al., 2005, 2009). This iterative
Figure 9. a) Template of 687 (semi)landmarks created on KNM-ER 406. b) Landmarks (black) and curve semilandmarks (grey) of the template. Labels are as in Table 2. c) Alignmentof OH5 by means of the estimated surface for the face (light grey) and neurocranium (dark grey). (def) Final reconstruction of OH5 from the left, top and fronto-lateral view.
S. Benazzi et al. / Journal of Human Evolution 61 (2011) 75e88 81
procedure approximates curves by sets of chords calculated asvectors of two neighboring semilandmarks, and surfaces by theirtriangulations. Once relaxed (namely, the bending energy betweencorresponding semilandmarks was minimized), semilandmarkscan be considered as geometrically homologous points. Missingdata are estimated freely in space by TPS interpolation (Gunz et al.,2005, 2009) from the information that is present; they are notrestricted to any curve or surface in the actual data.
The (semi)landmark configurations of the template and thetarget were imported into Amira software and its 104 (semi)
landmarks of the template were thus transformed into the corre-sponding (semi)landmarks of the target using the TPS interpolation(Bookstein, 1991), whereas the surface of the reference was warpedso as to minimize the bending energy of the according trans-formation. The resulting digital model was imported into Poly-Works, where the reconstructed part was isolated and “merged”with the upper fragment. For continuity between the temporal lineof the lower fragment and the reconstructed temporal line of theupper fragment (Fig. 5b, black spheres), an upward translation ofabout 1.5 mm was imposed, and also a concomitant backward
Table 2List of anatomical landmarks and curves of the template for OH5 reconstruction.
Unpaired landmarks Paired landmarks N Curve names Smlm counta
Basion (ba) Auriculare left (aul) 1 Basisphenoid 4Glabella (g) Auriculare right (aur) 2 Lower zygomatic arc left 14Inion (i) Condylus occipitalis posterior left (copl) 3 Lower zygomatic arc right 14Lambda (l) Condylus occipitalis posterior right (copr) 4 Median nuchal line 7Nasion (n) Frontomalare-orbitale left (fmol) 5 Midsagittal crest 14Opisthion (op) Frontomalare-orbitale right (fmor) 6 Orbital left 7Orale (ora) Frontomalare-temporale left (fmtl) 7 Orbital right 7Sphenobasion (sph) Frontomalare-temporale right (fmtr) 8 Palatine suture 7Staphilion (sta) Foramen ovale left (fovl) 9 Temporal line left 13
Foramen ovale right (fovr) 10 Temporal line right 13Zygotemporale inferior left (zygil) 11 Upper zygomatic arc left 14Zygotemporale inferior right (zygir) 12 Upper zygomatic arc right 14
Total semilandmarks on curves 128
a Semilandmarks identified on the curves.
S. Benazzi et al. / Journal of Human Evolution 61 (2011) 75e8882
translation along the y-axis to maintain continuity between thesurfaces of the two fragments in the right frontal process of thezygomatic bone (Fig. 5b, arrow).
The left frontal process of the maxillary bone was mirroredacross the midsagittal plane, restoring the right side (Fig. 5b). Sincethere are no known P. boisei specimens with a suitably preservedzygomatic and maxillary bones, CAD techniques were used forreconstructing the remaining missing region of the right hemiface.First, a curve network was digitized around the margin of themissing area (Fig. 5c). Second, a NURBS (non-uniform rationalbasis-spline) surface that matches the points along the curves(Piegl Les and Tiller, 1966; Piegl Les, 1991) was created in Rhino3.0(Robert MCNeel & Associates) (Fig. 5d). The NURBS surface wasconverted into a set of triangular faces (mesh) and “merged” to theOH5 face (Fig. 5e). The same procedure was used for the recon-struction of the posterior aspect of the frontotemporo-zygomaticregion. Small holes in themodel surfacewere filled automatically inPolyWorks. The ultimate symmetric reconstruction of the OH5 facewas accomplished by mirroring the reconstructed right hemifaceacross the midsagittal plane (Fig. 6a,b). Since a large portion of theleft hemiface was missing, thus reducing the amount of landmarksavailable, we avoided using the reflected relabeling approach (analternative method often used to restore symmetry) as it does notaccommodate missing data (Mardia et al., 2000).
Phase 3: OH5 neurocranial portion
The neurocranium needed to be symmetrized before we couldalign it with the face. The medio-posterior neurocranium is fairlywell represented, while the anterior part is completely missing(Fig. 7). The right side (Fig. 7a) is better preserved than the left(Fig. 7b). In particular, the left posterior temporal squama is missingand the pars mastoidea is almost completely lost. The zygomatic
Table 3Landmarks of the template for endocranial reconstruction.
Unpaired landmarks
Basion (ba) Auriculare left (aul)Glabella (g) Auriculare right (aur)Inion (i) Condylus occipitalis anterior leLambda (l) Condylus occipitalis anterior roNasion (n) Condylus occipitalis posterior lNasospinale (ns) Condylus occipitalis posterior rOpisthion (op) Foramen infraorbitale right (foOrale (ora) Frontomalare-orbitale left (fmoRhinion (rh) Frontomalare-orbitale right (fmStaphilion (sta) Frontomalare-temporale left (f
Frontomalare-temporale right
process is better preserved on the left side right up to the zygo-maticotemporal suture (Tobias, 1967) (Fig. 7b).
A simplemirroring of the right side would not provide a suitableresult. Because of the flaring of the crest, mirroring createsa “double crest” on both sides (Fig. 7c), which must be replaced bythe original one. Also, a step of about 2.3 mm was observedbetween the original bone and the mirrored one in the parietalbone due to the slight asymmetry of the neurocranium (Tobias,1967; Rak, 1983) (Fig. 7d). Accordingly, a template of 197 (semi)landmarks (11 landmarks and 186 semilandmarks) was built in theOH5 neurocranium (Fig. 8a, Table 1). In order to create a mirror/target configuration, the neurocranium was mirrored along itsmidsagittal plane while the (semi)landmarks were subjected toa reflected relabeling procedure. Semilandmarks needed to be slidover the surface of the mirrored neurocranium to positions that aregeometrically homologous with respect to the template configu-ration. Generalized Procrustes analysis (GPA) was performed toobtain a symmetric Procrustes average between the completedstructure and its reflected relabeling counterpart (Mardia et al.,2000). This involves translating and rescaling to a unit CentroidSize (CS), and rotating the configurations so that the distancesbetween corresponding (semi)landmarks areminimized (Rohlf andSlice, 1990; Bookstein, 1991). Procrustes coordinates of theProcrustes average configuration were transformed to Cartesiancoordinates by multiplying each (semi)landmark by the CS mean(themean between the CS of the two configurations). Following theprocedure mentioned above, in Amira software the templateconfiguration was used to warp the surface of the neurocraniumfrom the original configuration into the average one by TPS inter-polation (Bookstein, 1991).
The left hemi-neurocranium of the mean shape was substitutedby mirroring the right one. The flared midsagittal crest obtainedduring the mirroring process was replaced by the midsagittal crest
Paird landmarks
Foramen ovale left (fovl)Foramen ovale right (fovr)
ft (coal) Postglenoid left (pol)ght (coar) Postglenoid right (por)eft (copl) Zygomaxillare left (zygoml)ight (copr) Zygomaxillare right (zygomr)rir) Zygotemporale inferior left (zygil)l) Zygotemporale inferior right (zygir)or) P4eM1 (only left side)mtl) M1eM2 (only left side)(fmtr) M2eM3 (only left side)
Figure 10. a) Template of 331 (semi)landmarks created on KNM-ER 406 for the reconstruction of the OH5 endocranium. Labels are as in Table 3. b) Reconstruction of theendocranium.
Table 4List of OH5 versions.
Upper facialfragment
Matching theestimated surfaces
Neurocranium
2� clockwiseb 2� counterclockwiseb
Current position V1a V2 V32 mm downc V4 V5 V62 mm upc V7 V8 V9
a Version described along the paper.b Rotation about x-axis.c Translation in z-axis.
S. Benazzi et al. / Journal of Human Evolution 61 (2011) 75e88 83
of the mean shape (Fig. 8b). A final step was the restoration of thezygomatic arch that was lost with the entire left hemi-neuro-cranium during right hemi-neurocraniummirroring. This was donein RapidForm XOR2 by Iterative Closest Point (ICP) superposition(Besl and McKay, 1992; Zhang, 1994) of a copy of the originalstructures upon the now-symmetrized neurocranium, followed bya mirroring of its own (Fig. 8b).
Table 5Landmarks used in the template for Procrustes form-space analysis.
Unpair landmarks Pair landmarks
Basiona Auricularea I1I2c
Glabellab Canalis caroticus posteriora I2Cc
Iniona Condylus occipitalis anteriora CP3c
Lambdaa Condylus occipitalis posteriora P3P4c
Nasionb Frontomalare-orbitaleb P4eM1c
Opisthiona Frontomalare-temporaleb M1eM2c
Oralec Foramen ovalea M2eM3c
Prosthionc Postglenoida
Rhinionb Zygomaxillarec
Staphilionc Zygotemporale inferiorc
a Neurocranium.b Upper facial fragment.c Lower facial fragment.
Phase 4
The midsagittal plane of the neurocranium was aligned to themidsagittal plane of the face, which had already been aligned to theyz-plane of the Cartesian coordinate system. Consequently, thedegrees of freedom were reduced to three parameters of rigidmotion. If we agree with Tobias (1967) that the zygomatic processof the temporal bone should be placed in contact with the temporalprocess of the zygomatic bone at the zygomaticotemporal suture,translation in the y-axis of the neurocranium was constrained.Accordingly, the rotation axis was fixed at this contact area.
In order to deal with the two remaining degrees of freedom(translation in the z-axis and rotation about the x-axis), we firstextrapolated the missing cranial surfaces of the craniofacial
fragment and the neurocranium independently by warpinga reference cranium using iterative TPS (see below). Next, trans-lation in the z-axis and rotation about the x-axis of the neuro-cranium were carried out throughout the range within which theestimated surfaces did not overlap.
A CT scan of the cranium of KNM-ER 406 (P. boisei) wassegmented as mentioned above. On the reconstructed externalsurface, a template of 687 (semi)landmarks (21 landmarks, 128curve semilandmarks and 538 surface semilandmarks) was digi-tized (Fig. 9a,b; Table 2). In Edgewarp3D, the template was warpedboth onto the craniofacial portion (first target) and the neuro-cranium (second target) by iterative TPS.
In order to extrapolate the missing cranial surfaces of thecraniofacial fragment and neurocranium, the above mentioned(semi)landmarks configurations (reference and targets) wereimported in Amira software. Two final surface models wereobtained by warping the reference surface (KNM-ER 406) to thecraniofacial fragment and neurocranium, based on TPS interpola-tion of the 687 (semi)landmarks of the template and the two target
Figure 11. A spline was drawn between the greater wing of the left sphenoid (neurocranium) and the frontal bone (craniofacial fragment) in version V1, V7, V8 and V9, respectively.The arrow points toward the spline.
Figure 12. Procrustes form-space PCA in the space of the first three eigenvectorscomputed from HomoePanePongo, augmented by projections of our nine differentversions of OH5, Tobias’s version, Sts5 and KNM-ER 406; black ¼ V1; red ¼ V2eV6;dark grey ¼ V7eV9; light grey ¼ OH5t, brown ¼ KNM-ER 406; purple ¼ Sts5. This viewof the 3-axes plot is rotated to align PC1 with the long axis of the nine OH5 versions.
S. Benazzi et al. / Journal of Human Evolution 61 (2011) 75e8884
configurations, respectively. Considering the first model (referencewarped to the craniofacial fragment), the estimated surfaceimitates the original model in themedial part of the anterior cranialfossa (in proximity to the fracture), while it copies KNM-ER 406 inthe occipital region. The opposite is observed for the second model,in which the composite follows the data in the anterior part of themiddle cranial fossa while remaining a copy of the referencespecimen’s browridge. Because the gap between the face and theneurocranium is less than 3 cm, for the alignment of the two weonly needed to consider the first 3 cm of the estimated surface fromeach portion (Fig. 9c).
In order to align the two fragments of OH5, the position of theface was fixed while the neurocranium was allowed to translate inthe z-axis and rotate about the x-axis (defined at the zygomatico-temporal suture, Fig. 9c). The position that is visuallymost plausibleto us for the match of the two estimated surfaces was copied intothe final reconstruction, but other alternative solutions wereproduced in a range of �2� to assess the effect of this particularsubjective judgment.
After the relative position of the face with respect to thebraincasewas determined, the extrapolated surfaces were removedfrom both the face and braincase. There resulted a gap in the cranialvault that was reconstructed according to the same referencespecimen (KNM-ER 406) by using the face and the braincase asa single unit after their realignment.
Accordingly, in order to reconstruct this gap the referencespecimen was warped onto the aligned OH5. The area corre-sponding to the missing region in OH5 was isolated from theresulting warped surface and “merged” with the aligned OH5(Fig. 9def).
Endocranial reconstruction After the alignment of the OH5 faceand neurocranium, the resulting portion of the OH5 endocranialsurface was obtained as a negative mold of the internal neuro-cranium. We managed the entire reconstruction of the
endocranium by a single TPS interpolation, using KNM-ER 406 asthe reference specimen. The new template was made up of 31landmarks on the external surface (Table 3) in addition to 300surface semilandmarks digitized in the segmented endocranium
Figure 13. Superimposition between V1 (transparent grey) and OH5t (red). The arrows mark the principal differences between the two reconstructions.
S. Benazzi et al. / Journal of Human Evolution 61 (2011) 75e88 85
(Fig. 10a). We repeated the same processes as mentioned above forcreating the target configuration as well as for surface warping.
Phase 5: alternative OH5 versions
Our aimwas to reduce the remaining degrees of freedom for thealignment as far as possible based on explicit assumptions such assymmetry, smoothness of curvature or contact. We consider thefinal result (called V1 here) to be a particularly plausible version. Inorder to assess the range of uncertainty of our final product, wecreated eight further derivates of V1 and compared them with ourfinal reconstruction (V1) and the reconstruction provided by Tobias(1967). We do this in a typical context for shape analysis. Theseversions were prepared by perturbing two of the six rigid-bodydegrees of freedom. Starting from V1, two further versions wereobtained rotating the neurocranium relative to the face 2� clock-wise and 2� counterclockwise about the x-axis passing through thecontact between the parts at the zygomaticotemporal suture. Theupper facial fragment itself was then translated 2 mm downwardsor 2 mm upwards (translation in the z-axis), respectively, withconcomitant reappraisal of the contact between the two facialfragments in the frontal process of the right zygomatic bone(translation in the y-axis). For both of these facial alternatives, theneurocranium was rotated in three different positions: the first isthe visually most plausible fit matching the two estimated surfaces,which required a reassessment of the neurocranial position alongthe z-axis as well; the second and third versions rotated the neu-rocranium 2� clockwise or counterclockwise (Table 4).
Statistical analysis
A template of 44 anatomical landmarks (Table 5) and 643semilandmarks was created and warped by TPS onto a cranialsample of 24 modern humans, 21 Pan troglodytes and 18 Pongopygmaeus. CT data of the sample were acquired at the Radiologie 2Dept, Medizinische Universität Innsbruck, Austria, and at the RuberClinic Madrid, Spain, using a Siemens Somatom Plus 40 (Innsbruck)and a General Electric, model GE Light Speed 16 (Madrid). All CTscans were recorded in DICOM file format at a reconstructionmatrix size of 512 � 512 pixels. Pixel size ranged from 0.42 to0.51 mm and slice thickness from 0.625 to 1 mm.
The template was initially created on the cranium of Pongo, butwarped to the grand mean of the HomoePanePongo sample for useas template for the fossil specimens. Beside our OH5 reconstruc-tions V1eV9, a surface scan of the cast from Tobias’s OH5 recon-struction (OH5t), and CT scans of Sts5 and KNM-ER 406 wereincluded in the analysis. A principal component analysis (PCA) ofthe matrix of shape coordinates augmented by a column of thenatural logarithm of Centroid Size (lnCS), corresponding to a PCA inProcrustes form-space (Mitteroecker et al., 2004), was carried outon the original sample (HomoePanePongo) and all of our alterna-tive OH5 were projected into this space. The covariance matrix forthis analysis has to come from real data only, not from the recon-structed data, because our aim is to knowwhere they would plot asa fictional sample in an ordination of a real sample. Therefore, weproject them into the principal coordinate space of the otherspecimens. The intention is to demonstrate the range of variation in
S. Benazzi et al. / Journal of Human Evolution 61 (2011) 75e8886
Procrustes form-space spanned between some extant and extincthominoids for those reconstructions that arewithin plausible limitsof the given uncertainties. We prefer to show the range of variationof the nine alternative OH5 versions in Procrustes form-space,rather than shape-space, because size is so important an evolu-tionary factor. In form-space, PC1 usually captures most of theshape changes associated with lnCS (static allometry), while theother PCs represent various dimensions of shape variation usuallynot associated with size. For data processing and analyses, weused software routines written in R software (R Development CoreTeam, 2008).
Finally, in Rapidform XOR2 our final reconstruction V1 wasaligned to the right hemi-maxillary region and the lower portion ofthe right zygomatic bone of Tobias’s version by a best-fit point-to-surface alignment algorithm.
Results
The reconstructions obtained by an upward movement of theupper facial fragment (V7, V8, V9; Table 4) cannot be consideredplausible. As mentioned in Phase 2, the more the upper fragment isupwardly displaced, the more a backward translation is required tomaintain the continuity of the frontal process of the right zygo-matic bone. Since the neurocranium cannot move backward (theface and the neurocranium have a contact at the zygomatico-temporal suture), the reduction in space between face and neuro-cranium prevents a restoration of the missing part.
In order to document this problem, we have drawn a spline inV1, V7, V8, and V9 between the greater wing of the left sphenoid(belonging to the neurocranial portion) and the frontal bone of thecraniofacial fragment (Fig. 11). While in V1 there is still spaceenough between the left sphenoid and the frontal bone to allow fora smooth curvature, the shortened space characterizing the other
Table 6Anthropometric measurements (in millimeters) taken from Tobias on OH5t (1967)a com
a Information about and definitions of the measurements included in the table is provb Measurements made by Wood (1991) on a cast.c Linear measurements involving less than reliable terminus points are enclosed in sq
versions would demand a strong bending of the spline in anunrealistic way. Furthermore, the backward displacement of theupper facial fragment produces a depressed midfacial shape thatmakes it difficult to restore the continuity between the maxillarybone (lower fragment) and the frontal process of themaxilla (upperfragment). Accordingly, we can reject V7, V8, and V9 as plausiblealternatives for future quantitative comparison of form as well asfor studying dietary adaptations (however, we leave them in ourstatistical analysis to show where their form would plot incomparison).
Fig.12 is a selected projection of the first three dimensions of theProcrustes form-space PCA of the HomoePanePongo sample,including our range of reconstructed OH5. The first three PrincipalComponents (PCs) explain about 92%of the total variance, and so theordination here can be treated as fairly robust. The nine alternativeOH5 versions separate mostly along PC1, which is highly correlatedwith size (r ¼ 0.94). The uncertainties of reconstruction producespecific correlation between (reconstructed) centroid size and(reconstructed) shape that are different in their geometry from thesize allometry of the complete data from the three extant taxa. Inother words, there is an “allometry of reconstruction” that is thefocal topic of the display. In this view, rotated to alignwith the longaxis of the nine OH5 versions, all of our reconstructions appear to beabout the same distance from the Tobias’s version (OH5t), and theyare about the same distance from KNM-ER 406. Therefore, besidessizedifferences (PC1), thenineOH5versionsdifferalso in shape (PC2and PC3) from OH5t. So the difference of Tobias’s version from oursymmetric reconstructions induces the same correction of asym-metry, no matter which of the nine versions we use.
Naturally V1 is positioned near the middle of this reconstruc-tion’s point cloud. Tobias’s reconstruction (OH5t) plots at the samedistance as V3 and V6 regarding PC1 but has a higher score on PC2 inlight of its facial and cranial breadth reduction, forward browridge
S. Benazzi et al. / Journal of Human Evolution 61 (2011) 75e88 87
displacement and reduced midfacial protrusion. Also, OH5t differsfrom the reconstructions V1eV6 along PC3, owing to a forwardposition of its browridge with its elongated neurocranium.
The estimated endocranial capacity for V1 is 491 cm3. The rangefrom the smallest to the largest capacity of our reconstructionsV1eV6 is from 477 cm3 (V6) to 498 cm3 (V5) (for the full range of allnine derivates, this is from 470 cm3 (V9) to 498 cm3 (V5)). Much ofthe separation of OH5t from all of our reconstructions is due to theasymmetry of the Tobias’s version.
In Fig. 13, our final reconstruction V1 (rendered in transparentgrey) is superimposed to OH5t (rendered in red). The backwarddisplacement of both the left maxillo-zygomatic region and thezygomatic process of the frontal bone in OH5t (Fig. 13c) affects theposition of the entire neurocranium. In fact, the left zygomatic archis 7 mmmedially compressed (accounted by PC2 in Fig. 12; Fig. 13b)and upwardly positioned compared with V1 (Fig. 13c,d). On thecontrary, the zygomatic process of the right temporal bone is wellaligned with V1, even though backwardly displaced (Fig. 13d). Ingeneral, in view of the lack of constraints in the right zygomaticarch, the neurocranium of OH5t is more backwardly positionedthan V1 and turns toward the right side (Fig. 13b). Finally, Table 6present standard anthropometric measurements and indicesrecorded in V1 compared with those presented by other workers(e.g., Tobias, 1967; Wood, 1991).
Discussion
OH5 shows several disturbances of the original form. Thecranium is broken, several parts are missing, there are asymme-tries, and there is plaster and somematrix adhering to the fossilizedbones. Reconstruction is thus essential before the whole cranialform of this unique specimen can be analyzed and comparedquantitatively, for instance by applying coordinate-based methodssuch as geometric morphometric (GM). Our aimwas to provide thecompleted cranial geometry of P. boisei for exploring hypotheses ofthe mechanics of early hominin feeding and diets.
Such biomechanical analysis via Finite Element Modeling (FEM)cannot work with (distorted) fragments, but needs the entiresurface geometry of the restored cranium (cf. Strait et al., 2009).
Our work is a derivative of Tobias’s (1967) earlier reconstruc-tion, which was conducted under the much more challengingcircumstances prevailing at the time. Currently, virtual recon-struction allows one to manipulate the fragments in a gravity-freeenvironment, to control all steps by numbers, to remove and addmaterial without affecting the original, and forces the recon-structor to make considerations and assumptions explicit. Thedegrees of freedom for our repositioning of fragments werestrongly reduced by introducing individual midsagittal planes.Moreover, the reference data from another P. boisei cranium notavailable to Tobias (1967) facilitated the estimation of thosemissing surfaces. We have shown our reconstruction steps andpossible alternatives here in a detailed manner that is rare in theliterature (Weber and Bookstein, 2010). As a consequence, theuncertainties left for the positioning are within a range of 1e2 mmor 1e2�, and we provided some alternatives that are possiblewithin these narrow limits. Some of them are too extreme and donot allow a reasonable anatomical reconstruction of OH5 (V7, V8and V9). However, the other options are plausible (from V1 to V6).Their range of variability was shown in Fig. 13, and we cannot claimto define the right reconstruction. Nevertheless, two of us (SB &GW) agreed that V1 is the most plausible reconstruction. Theendocranial capacity of V1 is 491 cm3, somewhat less than Tobias’estimation of about 530 cm3, and closer to the 500 cm3 estimatedby Falk et al. (2000). We suggest using this version for furtheranalysis (the digital data will be made available for other
researchers after the termination of the project in 2012). Oneimportant validation of the reconstruction procedure comes fromthe perfect matching of the face and neurocranium at the zygo-maticotemporal suture. Notably, the face and neurocranium weresymmetrized independently (and using different processes) andwere only then aligned using the midsagittal planes.
Earlier OH5 reconstructions are less symmetrical than ourvirtual reconstructions. As pointed out by Tobias (1967) anddemonstrated by Rak (1983), OH5 itself has anatomic elements thatare genuinely unsymmetrical, like the rugosity and shape of thesagittal crest and the uneven wear on the dentition. We are awarethat symmetry, or its opposite, asymmetry, is a significant aspectfor biomechanical analyses. For this reason, we used the asym-metrical sagittal crest in our virtual model. Nevertheless, we alsoemphasize that obvious post-mortem deformations (such as theskewed upper face) need a correction prior to further analysis.
Conclusions
From the immediate biological point of view, this reconstructiondoes not result in breaking newse it does not question Tobias’s andother workers’ general thoughts on the hypodigm of P. boisei e andthis is, after all, not what a reconstruction per se is likely to produce.This virtual reconstruction of OH5 is nothing more and nothing lessthan the careful restoration of its cranial geometry using state-of-the-art methods, the preserved anatomy, and some very generalknowledge about mammalian crania. The result, however, can nowbe used for testing further hypotheses (i.e., for studying dietaryadaptations via finite element analysis e FEA), and thus leads tonew insights.
Acknowledgments
We thank Michael Coquerelle and Sascha Senck for their helpand valuable advice. Wewish to express our thanks to Philipp Gunzfor providing the 3D surface scan of Tobias’s reconstruction. Thiswork was supported by the National Science Foundation PhysicalAnthropology HOMINID program (NSF BCS0725219, 0725183,0725147, 0725141, 0725136, 0725126, 0725122, 0725078) and the"European Union FP6 Marie Curie Actions MRTN-CT-2005e019564“EVAN”.
References
Besl, P.J., McKay, N.D., 1992. A method for registration of 3-d shapes. IEEE Trans.Pattern Anal. Mach. Intell. 14, 239e256.
Bookstein, F.L., 1991. Morphometric Tools for Landmark Data. Geometry andBiology. Cambridge University Press, New York.
Bookstein, F.L., Green, W.K., 2002. Edgewarp 3D Available from: ftp://brainmap.stat.washington.edu/pub/fred/ewsh3.19.man/.
Falk, D., Redmond, J.C., Guyer, J., Conroy, C., Recheis, W., Weber, G.W., Seidler, H.,2000. Early hominid brain evolution: a new look at old endocasts. J. Hum. Evol.38, 695e717.
Gunz, P., Mitteroecker, P., Bookstein, F.L., 2005. Semilandmarks in three dimensions.In: Slice, D.E. (Ed.), Modern Morphometrics in Physical Anthropology. KluwerAcademic/Plenum Publishers, New York, pp. 73e98.
Gunz, P., Mitteroecker, P., Neubauer, S., Weber, G.W., Bookstein, F.L., 2009. Principlesfor the virtual reconstruction of hominin crania. J. Hum. Evol. 57, 48e62.
Hay, R.L., Kyser, T.K., 2001. Chemical sedimentology and paleoenvironmental historyof Lake Olduvai, a Pliocene lake in northern Tanzania. Bull. Geol. Soc. Am. 133,1505e1521.
Kalvin, A.D., Dean, D., Hublin, J.J., 1995. Reconstruction of human fossils. Commu-niqué 3, 4e6.
Leakey, L.S.B., 1959. A new fossil skull from Olduvai. Nature 201, 967e970.Mardia, K.V., Bookstein, F.L., Moreton, I.J., 2000. Statistical assessment of bilateral
symmetry of shapes. Biometrika 87, 285e300.Mitteroecker, P., Gunz, P., Bernhard, M., Schaefer, K., Bookstein, F.L., 2004.
Comparison of cranial ontogenetic trajectories among great apes and humans.J. Hum. Evol. 46, 679e698.
S. Benazzi et al. / Journal of Human Evolution 61 (2011) 75e8888
Neeser, R., Ackermann, R.R., Gain, J., 2009. Comparing the accuracy and precision ofthree techniques used for estimating missing landmarks when reconstructingfossil hominin crania. Am. J. Phys. Anthropol. 140, 1e18.
Neubauer, S., Gunz, P., Weber, G.W., 2005. Digital reconstruction of P. boisei OH5.Am. J. Phys. Anthropol. 126 (S 40), 156.
Ogihara, N., Nakatsukasa, M., Nakano, Y., Ishida, H., 2006. Computerized restorationof nonhomogeneous deformation of a fossil cranium based on bilateralsymmetry. Am. J. Phys. Anthropol. 130, 1e9.
O’Higgins, P., Cobb, S.N., Fitton, L.C., Gröning, F., Phillips, R., Liu, J., Fagan, M.J., 2010.Combining geometric morphometrics and functional simulation: an emergingtoolkit for virtual functional analyses. J. Anat. doi:10.1111/j.1469-7580.2010.01301.x.
Piegl Les, A., 1991. On NURBS: a survey. IEEE Comput. Graph. 11, 55e71.Piegl Les, A., Tiller, W., 1966. The NURBS Book. SpringereVerlag, New York.Rak, Y., 1983. The Australopithecine Face. Academic Press, New York.R Development Core Team, 2008. R: A Language and Environment for Statistical
Computing. R Foundation for Statistical Computing, Vienna, Austria, ISBN 3-900051-07-0. http://www.r-project.org.
Rohlf, F.J., Slice, D., 1990. Extensions of the procrustes method for the optimalsuperimposition of landmarks. Syst. Zool. 39, 40e59.
Schwartz, J.H., Tattersall, I., 2005. The Human Fossil Record, CraniodentalMorphology of Early Hominids (Genera Australopithecus, Paranthropus, Orrorin),and Overview, vol. 4. WileyeLiss, New York.
Grosse, I., Byron, C., Wood, B.A., Lawn, B., Constantino, P., Slice, D.E., Smith, A.L.,2009. The feeding biomechanics and dietary ecology of Australopithecus afri-canus. Proc. Natl. Acad. Sci. 106, 2124e2129.
Tamrat, E., Thouveny, N., Taieb, M., Opdyke, N.D., 1995. Revised magneto-stratigraphy of the PlioePleistocene sedimentary sequence of the Olduvaiformation (Tanzania). Palaeogeogr. Palaeoclimatol. Palaeoecol. 114, 273e283.
Tobias, P.V., 1967. Olduvai Gorge Vol. 2: the Cranium and Maxillary Dentitionof Australopithecus (Zinjanthropus) boisei. Cambridge University Press,Cambridge.
Weber, G.W., Bookstein, F.L., 2010. Virtual Anthropology: A Guide to a New Inter-disciplinary Field. SpringereVerlag, Wien, New York.
Wolpoff, M.H., 1999. Paleoanthropology, second ed. McGraw Hill, New York.Wood, B.A., 1991. Koobi Fora Research Project. Hominid Cranial Remains, vol. 4.
Clarendon Press, Oxford.Wood, B., Costantino, P., 2007. Paranthropus boisei: fifty years of evidence and
analysis. Yearb. Phys. Anthropol. 50, 106e132.Wroe, S., Ferrara, T.L., McHenry, C.R., Curnoe, D., Chamoli, U., 2010. The cranio-
mandibular mechanics of being human. Proc. R. Soc. Lond. B. doi:10.1098/rspb.2010.0509.
Zhang, Z.Y., 1994. Iterative point matching for registration of free-form curves andsurfaces. Int. J. Comput. Vision. 13, 119e152.
Zollikofer, C.P.E., Ponce De León, M.S., 2005. Virtual Reconstruction: A Primer inComputer-Assisted Paleontology and Biomedicine. Wiley Interscience, Hobo-ken, New Jersey.
