Three-dimensional geometric morphometrics of thorax-pelvis covariation and its potential for predicting the thorax morphology: A case study on Kebara 2 Neandertal

Abstract

The skeletal torso is a complex structure of outstanding importance in understanding human body shape evolution, but reconstruction usually entails an element of subjectivity as researchers apply their own anatomical expertise to the process. Among different fossil reconstruction methods, 3D geometric morphometric techniques have been increasingly used in the last decades. Two-block partial least squares analysis has shown great potential for predicting missing elements by exploiting the covariation between two structures (blocks) in a reference sample: one block can be predicted from the other one based on the strength of covariation between blocks. The first aim of this study is to test whether this predictive approach can be used for predicting thorax morphologies from pelvis morphologies within adult Homo sapiens reference samples with known covariation between the thorax and the pelvis. The second aim is to apply this method to Kebara 2 Neandertal (Israel, ∼60 ka) to predict its thorax morphology using two different pelvis reconstructions as predictors. We measured 134 true landmarks, 720 curve semilandmarks, and 160 surface semilandmarks on 60 3D virtual torso models segmented from CT scans. We conducted three two-block partial least squares analyses between the thorax (block 1) and the pelvis (block 2) based on the H. sapiens reference samples after performing generalized Procrustes superimposition on each block separately. Comparisons of these predictions in full shape space by means of Procrustes distances show that the male-only predictive model yields the most reliable predictions within modern humans. In addition, Kebara 2 thorax predictions based on this model concur with the thorax morphology proposed for Neandertals. The method presented here does not aim to replace other techniques, but to rather complement them through quantitative prediction of a virtual ‘scaffold’ to articulate the thoracic fossil elements, thus extending the potential of missing data estimation beyond the methods proposed in previous works.

Introduction

The skeletal torso is a complex structure of considerable importance in understanding the evolution of human body shape and its paleobiological implications (Ruff, 1991; Schmid, 1991; Carretero et al., 2004; Holliday, 2012; Jungers et al., 2016; Bastir et al., 2020). Torso reconstruction in fossil hominins is crucial to shed light on their body plans, particularly in view of how tightly it is linked to body mass and stature (Ruff et al., 1997; Arsuaga et al., 1999, 2015; Holliday, 2012; Carretero et al., 2004; Simpson et al., 2008; Ruff, 2010), and potentially body energetic demands (Franciscus and Churchill, 2002; Churchill, 2006, 2014; Froehle and Churchill, 2009; García-Martínez et al., 2018a, b, c; Gómez-Olivencia et al., 2018). However, torso reconstruction may pose challenges because its morphology depends on the interaction not only within the thorax (ribs, vertebrae, and sternum), the lumbar spine, and the pelvis but also between them and their associated soft tissues, which are not generally preserved in the fossil record. Fortunately, as thorax and pelvis morphologies covary (Middleton, 2015; Torres-Tamayo et al., 2018), these anatomical parts can inform us about each other, and previous researchers used this information to elucidate complete hominin torso morphologies (Schmid, 1983, 1991; Jellema et al., 1993; Walker and Ruff, 1993; Tague and Lovejoy, 1998; Sawyer and Maley, 2005; Simpson et al., 2008; Berge and Goularas, 2010; Schmid et al., 2013; Latimer et al., 2016; Brassey et al., 2018; Gómez-Olivencia et al., 2018; Laudicina et al., 2019).

In some of the aforementioned studies, reconstructions of skeletal trunk elements have been made through classical methodological approaches using different materials, such as wax or plaster (e.g., Schmid, 1983; Walker and Ruff, 1993; Sawyer and Maley, 2005; Simpson et al., 2008). However, virtual reconstruction methods have become increasingly popular in the past few years and are an excellent alternative to these classic techniques as they reduce the risk of damaging the original fossils during handling (Zollikofer and Ponce de León, 2005; Gunz et al., 2009; Weber and Bookstein, 2011; Bastir et al., 2019a). Among these techniques, geometric morphometrics (GM) has become an important tool for aiding in fossil reconstruction, and a 3D approach, in addition to the inclusion of sliding semilandmarks, has considerably improved the estimation of missing data (Slice, 2005, 2007; Benazzi et al., 2009; Gunz et al., 2009; Mitteroecker and Gunz, 2009; O'Higgins et al., 2011; Gunz and Mitteroecker, 2013; García-Martínez et al., 2014; Brassey et al., 2018; Schlager et al., 2018). In turn, these methods have been the basis for virtual reconstructions of different trunk fossil elements (Berge and Goularas, 2010; García-Martínez et al., 2014, 2018d; Claxton et al., 2016; Brassey et al., 2018; Rmoutilová et al., 2019).

Among the missing data estimation techniques are those using multiple multivariate regressions for statistical reconstruction based on a reference sample (Bookstein et al., 2003; Gunz et al., 2004; Weber and Bookstein, 2011; Stelzer et al., 2018). In this approach, multiple variables (e.g., a set of 3D coordinates on an anatomical structure) are regressed on all other variables in a reference sample of complete specimens, and missing values (3D coordinates) are predicted by the generated linear regression model. Gunz et al. (2004) presented this method for reconstructing incomplete human crania and demonstrated that regression-based reconstruction was more accurate than the thin plate spline warping and mean substitution methods. This technique has been recently used for fossil reconstruction by Stelzer et al. (2018), who demonstrated that dental arcades of extinct hominins can be reliably predicted using the covariation between upper jaws and lower jaws of a reference sample composed of extant hominoids with previously demonstrated known associations (Stelzer et al., 2017).

The two-block partial least squares (2B-PLS) regression method was first applied to investigate covariation in shape data by Rohlf and Corti (2000), and subsequent studies have used this method to statistically assess the covariation between two or more different sets of shape variables (Bookstein et al., 2003; Bastir et al., 2005; Mitteroecker and Bookstein, 2007; Mitteroecker et al., 2012; Klingenberg and Marugán-Lobón, 2013; Adams and Collyer, 2016; Arlegi et al., 2018; Neaux et al., 2018; Scott et al., 2018; Torres-Tamayo et al., 2018). Gunz et al. (2009) referred to the 2B-PLS analysis as a method to reduce high dimensionality of data in regression-based predictive analysis while exploiting the morphological integration between the known and missing parts. Furthermore, it has also shown great potential for predicting missing elements: one block can be predicted from the other one based on the strength of covariation between blocks in a reference sample (Schlager, 2013; Archer et al., 2018; Torres-Tamayo et al., 2019; Bastir et al., 2019b). By studying stone tools, Archer et al. (2018) demonstrated that a flake body could be accurately predicted from the platform body based on the covariation between these two structures. In testing whether this predictive analysis could be applied to an organism, Bastir et al. (2019b) successfully predicted anatomically connected lumbar spines from isolated lumbar vertebrae in modern humans, thus validating this method for the first time in an anatomical system.

One of the best fossil examples to which classic and virtual techniques have been applied is the Kebara 2 specimen (Homo neanderthalensis, ∼60 ka; Valladas and Valladas, 1991). This well-preserved young adult, presumed to be a male, was found in Kebara Cave (Israel) in 1983 and has been used as a key specimen in describing Neandertal postcranial anatomy (Rak and Arensburg, 1987; Vandermeersch, 1991; Been et al., 2010; Gómez-Olivencia et al., 2009, 2013, 2017, 2018; García-Martínez et al., 2014; Chapman et al., 2017). The thorax, which is usually badly preserved mainly due to the fragility of the ribs, has been the focus of several works that characterized isolated thoracic elements of Kebara 2 and helped to lay the foundation to describe the thoracic morphology of this specimen (Arensburg, 1991; Gómez-Olivencia et al., 2009; García-Martínez et al., 2014, 2018a, b, c; Been et al., 2017; Chapman et al., 2017).

The first approach to a Neandertal skeleton reconstruction was performed by Sawyer and Maley (2005), who based their reconstruction of La Ferrassie 1 on recovered material from other Neandertal skeletons. Among these specimens, these authors used the thoracic vertebrae, ribs, and sternum of Kebara 2 Neandertal to reconstruct the thorax, using clay and epoxy paste to fill missing areas. This thorax exhibits a dome-shaped upper part and a markedly flaring lower part, which, together with notable rib declination, results in an anteroposterior flattened bell-shaped reconstruction. According to Sawyer and Maley (2005: 30), “Although the ribcage and pelvis are visually compelling and convincing in demonstrating the relative difference in Neandertals and modern humans, the introduction of some degree of artistic license makes it difficult to comment on the significance of these differences.” As part of this artistic license, the authors altered the first and second rib lengths of their thorax reconstruction to better match the somewhat larger La Ferrassie 1 shoulder girdle. Even so, the bell-shaped thorax reconstruction performed by Sawyer and Maley (2005) has been considered the only approach to Kebara 2 thorax morphology until recently.

Gómez-Olivencia et al. (2018) combined published data and anatomical expertise to reconstruct the thorax of Kebara 2 using 3D virtual techniques. These authors first reconstructed the thoracic vertebral column of this specimen by slightly modifying a previous reconstruction undertaken by Been et al. (2017), then correcting taphonomic and reconstruction deformation of the isolated ribs and using mirror images when necessary, and virtually articulating the ribs and sternal elements. This reconstruction was analyzed using both traditional morphometrics and 3D geometric morphometrics (3DGM) within a comparative framework composed of 16 thoraces of adult male Homo sapiens individuals. These authors found that Kebara 2 showed a wider lower thorax with a pronounced invagination of the thoracic spine and a more horizontal orientation of the ribs than modern human males of similar stature.

The reconstructions produced by Sawyer and Maley (2005) and Gómez-Olivencia et al. (2018) reinforce the widely accepted morphology of the Neandertal thorax: this species showed a large costal skeleton, especially in the mid-thoracic ribs, and more dorsally oriented transverse processes in the mid-thoracic spine that leads to a wider lower thorax in the mid-lower segment (Franciscus and Churchill, 2002; Sawyer and Maley, 2005; Gómez-Olivencia et al., 2009, 2018, 2019; García-Martínez et al., 2014, 2017, 2018a, b, c; Bastir et al., 2015, 2017; Gómez-Olivencia et al., 2018, 2019). However, achieving these reconstructions has proved difficult as the thorax is a complex structure composed of different elements that need to be articulated, and this requires vast anatomical expertise. In vivo, these structures are anatomically connected through the costovertebral and the costochondral joints (Graeber and Nazim, 2007; Beyer et al., 2014), so thorax morphology relies on anatomical relationships between these elements. However, these cartilaginous joints are generally not preserved in the fossil record, and thus, any thorax reconstruction relies on how ribs are articulated to their corresponding thoracic vertebrae. As a consequence, there could be more than one way to align the thoracic elements depending on each researcher's criteria, which could affect rib orientation and declination, thus modifying the anteroposterior, mediolateral, and craniocaudal diameters of the articulated thorax. These different assumptions of researchers might also be based on different interpretations of the body shapes of extinct hominins, which is then reflected in their reconstructions (Schmid, 1983; Bonmatí et al., 2010; Arsuaga et al., 2015; Latimer et al., 2016; Gómez-Olivencia et al., 2018; Haeusler et al., 2019). This reinforces the necessity of developing new and complementary approaches for such reconstructions.

This work aims to test whether thorax morphologies can be predicted from pelvic morphologies using 3DGM techniques and 2B-PLS analysis, working as a predictive method (Schlager, 2013; Archer et al., 2018; Torres-Tamayo et al., 2019; Bastir et al., 2019b). This method was first tested on a comparative sample composed of living humans with known associations between thorax and pelvis shape and then applied to the Kebara 2 Neandertal specimen, the thoracic and pelvic morphologies of which are well known, thanks to previous reconstructions of these two anatomical structures (Rak and Arensburg, 1987; Sawyer and Maley, 2005; Gómez-Olivencia et al., 2018). We used two different pelvic predictors and three different thoracopelvic covariation models based on an adult living H. sapiens reference sample. The resulting six thorax predictions were compared with each other and with the two thorax reconstructions of Kebara 2 Neandertal previously published (Sawyer and Maley, 2005; Gómez-Olivencia et al., 2018). Finally, the limitations of this predictive method are discussed.