Anatomical networks reveal the musculoskeletal modularity of the human head

Mosaic evolution is a key mechanism that promotes robustness and evolvability in living beings. For the human head, to have a modular organization would imply that each phenotypic module could grow and function semi-independently. Delimiting the boundaries of head modules, and even assessing their existence, is essential to understand human evolution. Here we provide the first study of the human head using anatomical network analysis (AnNA), offering the most complete overview of the modularity of the head to date. Our analysis integrates the many biological dependences that tie hard and soft tissues together, arising as a consequence of development, growth, stresses and loads, and motion. We created an anatomical network model of the human head, where nodes represent anatomical units and links represent their physical articulations. The analysis of the human head network uncovers the presence of 10 musculoskeletal modules, deep-rooted in these biological dependences, of developmental and evolutionary significance. In sum, this study uncovers new anatomical and functional modules of the human head using a novel quantitative method that enables a more comprehensive understanding of the evolutionary anatomy of our lineage, including the evolution of facial expression and facial asymmetry.

most cases 'functional components' are merely informed conjectures based on assumptions about position, form and function that reflect a priori expectations rather than the results of quantitative analysis 14,15 . For example, a recent anatomical network analysis (AnNA) showed that functional matrices are essential to generate a proper pattern of connectivity of the face, whereas the pressure of the growing brain against the skull vault is not necessary to explain the connectivity pattern among the bones of the cranial vault and base 15 .
Here we use this new quantitative and objective approach (AnNA) to treat the skeletal, cartilaginous, and muscular units of the human head as the elements of a network (nodes), whose interactions at their physical contacts (links) determine the boundaries of the phenotypic modules of the head ( Fig. 1 and Table 1; see Methods and SI for further details). Our driving hypothesis was that we should be able to (1) define phenotypic modules that reflect developmental, functional, and morphological aspects of the anatomy of the head, and thus (2) identify at least some modules that differ from those that were predicted purely by a priori theoretical or qualitative assumptions. Using AnNA also allowed us to analyze bone dependences in isolation from muscle dependences to further enrich our understanding of human head modularity.
Our study revealed that the musculoskeletal network of the adult human head and neck comprises 181 morphofunctional units (bones, cartilages and muscles) connected by 412 physical contacts. The head divides into 10 musculoskeletal modules that form coherent anatomical, functional, evolutionary and/or developmental complexes, which have never been suggested in the past ( Fig. 1 and Table 1). Thus, studying general biological dependences using this well-defined quantitative method (AnNA) reveals unique insights about human head complexity-specifically its development, evolutionary origins and diseases-that are not readily apparent using conventional approaches.
The lower jaw/inner ear musculoskeletal complex (module 1) is particularly interesting because it groups morphofunctional units that would intuitively seem independent from each other given their anatomical contacts. Remarkably, AnNA highlights an unexpected, deeper connection: these units are linked by structures that share a major common developmental denominator, the first pharyngeal arch. For instance, this module comprises neurocranial bones and facial muscles of the ear region (e.g. auricularis posterior), which are not themselves derived from the first arch but contact a bone (malleus) and a muscle (tensor tympani) that are first arch derivatives. The skull area surrounding the ear region is in turn connected to the lower jaw by first arch muscles, such as the masseter, temporalis, pterygoideus lateralis, and digastricus anterior (via the digastricus posterior, which is a second arch muscle). Further, the mandible is connected via other first arch muscles (e.g. mylohyoideus) to the hyoid bone, which is a second arch structure; and thus also to the tongue, infrahyoid muscles, and some pharyngeal muscles. Evolutionarily, this musculoskeletal complex is particularly interesting because it reveals an intricate interplay between an ancient relationship in mammals (i.e. the lower jaw and inner ear bones) and various muscles that originated before the rise of mammals (e.g. genioglossus, geniohyoideus), alongside masticatory muscles with clear non-mammalian homologues 16,17 .
The mid/upper face musculoskeletal complex (module 2), which groups upper facial bones and muscles, illustrates how AnNA can coherently synthesize data from different sources (i.e. origin, growth, and function) to detect phenotypic modules not predicted using theoretical assumptions. As explained above, in Moss' model, the temporalis, masseter, and medial pterygoid muscles were grouped into a single module. However, AnNA groups the medial pterygoid muscle in the mid/upper face musculoskeletal module, and the masseter with the temporalis in the lower jaw/inner ear musculoskeletal module. Significantly, studies of human development pathologies (e.g. cleft lip and palate) have consistently shown a strong developmental and functional relationship between the upper and midface muscles salient to facial expression and musculoskeletal units related to palate movements 18 . Our results further support the idea that integrating muscle as well as skeletal modules yields new and deeper insights relevant in evolutionary developmental and medical contexts 10 .
The laryngeal musculoskeletal complex (module 3) constitutes a well-defined phenotypic module that includes the laryngeal cartilages and the muscles directly attaching these cartilages. The neck musculoskeletal complex (module 4) includes all the neck muscles innervated by cranial nerves that attach the skull to the nearby postcranial bones (i.e. cervical vertebrae, clavicles, scapulae, and sternum). This neck module is interesting as it groups muscles and bones with completely different developmental and evolutionary origins, indicating that this module is mainly defined by function. The left and right oral/ocular complexes (modules 5 and 6) group the maxillae and the zygomatic bones with orofacial muscles (see the left and right orofacial muscular complexes described below) together with the zygomaticus minor (orofacial), the depressor supercilii (ocular), and the inferior oblique (extrinsic) muscles. The left and right superficial ear complexes (modules 7 and 8) and the left and right inner ear complexes (modules 9 and 10) are also coherent functional modules: the former include only facial muscles related to the movements of the ear, the latter include only inner musculoskeletal structures of the inner ear. It is interesting to note that in the network analysis including only muscles the zygomaticus minor -an elevator of the upper lip -is not included in the orofacial muscle module with the zygomaticus major, while these two muscles are grouped in a same module in the network analysis including both muscles and the skeleton, as would be expected a priori based on function.
Importantly, the use of AnNA also allows one to efficiently separate the musculoskeletal network into its two main component networksone skeletal and one muscular-thus facilitating the independent analysis of hard and soft morphofunctional units. The skeletal network comprises 45 bones and cartilages articulated at 86 contact surfaces (sutures, synchondroses, and synovial joints). This skeletal network divides into eight modules, which are shown in Fig. 2 and in Table 1. Among these eight well-delimited modules are a cranial (neurocranium and basicranium) and a facial (viscerocranium) complex as previous studies have reported 9 , thus indicating that AnNA can detect and further validate accepted modules. A further strength of the present work is that it is the first AnNA study to also include the mandible, the ear ossicles, the hyoid bone, and the laryngeal cartilages. By doing this, this study reveals that the cranial module includes the mandible with the bones of the vault and cranial base, because of the mandible's structural relation with the temporal bones (e.g. glenoid fossa); in contrast, the left and right ossicles complexes group auditory ossicles independently from other bones. The thyroid complex groups all laryngeal cartilages, while the hyoid bone forms its own module (by not including muscles in the skeletal networks, the hyoid bone is not connected directly to others skeletal structures). We included vertebrae, sternum, scapulae, and clavicles in our analysis because these bones also connect with head muscles (e.g. trapezius, sternocleidomastoideus, platysma): AnNA grouped them in two separate modules, the thoracic and the cervical complexes, because they are isolated by the absence of muscular attachments. In turn, the muscular network comprises 136 muscles sparsely connected at 78 contact points (fiber fusions and well-defined tendons), and divides into three major modules and 21 smaller blocks of 4 to 2 muscles each. The three main modules are shown in Fig. 3 and in Table 1: a single ocular/upper face complex, and left and right orofacial complexes. It is remarkable that the three main muscular modules include muscles of facial expression exclusively. Recent comparative studies of primates have shown that facial expression muscles have undergone more evolutionary change (e.g. in shape, in appearance and loss, and in insertion shifting) than most other groups of head muscles during human evolution 17,19,20 . In addition, the evolution of facial muscles has been crucial to our particular abilities for verbal and visual communication 21 . Interestingly, none of these major and minor muscular complexes derive from a shared ontogenetic anlage, or a homogeneous developmental origin. For instance, some modules group a muscle of the 1 st arch (e.g. digastricus anterior) with muscles of the 2 nd arch (e.g. digastricus posterior, stylohyoideus) rather than other muscles of the 1 st arch. Instead, and importantly, muscular modules are functional complexes that integrate muscles with completely different phylogenetic and developmental origins.
Further, our results bring new light to the debate on the symmetry/ asymmetry of facial expression muscles in humans and primates 22,23 . Recent developmental studies suggest that the left and right facial muscles separate from each other early in ontogeny: but in fact, the left muscles are actually ontogenetically more closely related to the base of the pulmonary trunk, and the right ones to the base of the aorta 24 . Also, functional studies in humans show that asymmetrical use of facial muscles is crucial to make complex facial expressions 25 . Furthermore, functional and anatomical studies of human facial expressions have shown that asymmetrical use of facial muscles is less prominent, and that innervations patterns of muscles are more symmetric, in the upper face (muscles located above the upper brow) than in the mid-face and lower face 26,27 . Since human speech tends to involve symmetrical muscle contraction, asymmetrical use of facial muscles is likely related to non-verbal communication in our own species. The phenotypic modules identified here place these developmental, functional, and anatomical observations in a completely new and quantitative context: contrary to expected bilateral orofacial muscular and musculoskeletal complexes, here we report the presence of left and right orofacial modules. This supports the ontogenetic separation of left and right facial muscles and the ability to asymmetrically contract or relax facial muscles, and thus strike more complex facial expressions in humans. In addition, AnNA recovered a single module including both the left and right ocular/upper face facial muscles, in line with previous studies showing that innervations patterns and use of muscles are more symmetric in the upper face. Future studies will lead us to apply AnNA specifically to muscles of facial expression among other primate and mammal species to investigate which anatomical structures may be unique to humans and which others have deeper evolutionary origins.  Anatomical network modeling. We built an anatomical network model of the head's musculoskeletal system, which comprises all anatomical units of the human head, as well as the different types of physical interaction among them (Supplementary Data). For the purpose of this study, the skeletal and the muscular systems were analyzed separately-as two independent network models-in addition to the analysis carried out for the network model representing the entire musculoskeletal system of the head. Thus, we used different definitions of node and connection for each of these network models. The skeletal network comprises the bones and cartilage of the head and associated structures (skull, ear ossicles, mandible, neck cartilages, cervical vertebrae, and upper thoracic bones): nodes represent bones and connections represent physical articulations among them (sutural, synchondrosal, and synovial). The muscular network comprises the muscles of the head: nodes represent muscles and connections represent tendinous joints and fibrous fusions among them. The musculoskeletal network comprises all the above-mentioned anatomical parts of the head: nodes represent bones, cartilages, and muscles, and connections represent the abovedescribed physical articulations, as well as fibrous, and tendinous attachments of muscles onto bones and cartilages. Network nodes were coded in and stored as igraph objects using the igraph package in R 28 .

Methods
Identifying connectivity modules in musculoskeletal networks. A connectivity module is here defined as a group of anatomical units with more connections among them than to other units outside their group 3 . We identified the number and composition of connectivity for each anatomical network by maximizing the strength of modularity quantified as the modularity Q-value over all potential partitions 29 . We identified potential partitions in the musculoskeletal network using an heuristic method: first we performed a walk-trap algorithm of length 3 and then we resolved the best partition by taking the division that outputs the maximum Q value 30 . Q is the difference between the actual proportion of the connections within nodes in the same module and the expected proportion in a random network Q 5 P M (e mma m 2 ), where M is the total number of modules, e mm is the proportion of links within module m, and a i is the proportion of links of nodes in m. Q ranges from 21 to 1: Q . 0 indicates that the number of the connections among elements within the same module are higher than expected at random. In networks with a significantly strong modular organization Q varies from 0.3 to 0.7, higher values being rare 29 . The identification of modules in anatomical networks was performed using the igraph package in R 28 .