Cell–cell signaling through fibroblast growth factors and their receptors (FGF/FGFR) plays fundamental roles in development (Montero et al., 2000; Ornitz, 2005) and has been proposed as a key system in the evolution of the vertebrate head (Bertrand et al., 2011). The FGF/FGFR signaling system is implicated in the control of basic processes (e.g., proliferation, differentiation, migration, polarity, adhesion, apoptosis) in cells destined to become various tissues including bone and cartilage (Ornitz and Itoh, 2001; Dorey and Amaya, 2010; Li et al., 2013). Mutations in FGFR1, 2 and 3 can lead to varied impairments of skeletal development as demonstrated by the FGFR-related craniosynostosis syndromes (i.e., Apert [AS], Beare-Stevenson, Crouzon [CS], Crouzon with acanthosis nigricans, Jackson-Weiss, Muenke [MS], and Pfeiffer [PS] syndromes). These syndromes are characterized by premature fusion of one or several cranial vault sutures associated with skull dysmorphology and potentially presenting with malformations affecting the limbs, upper airway, brain, spine, heart, and/ or lungs (Cohen and MacLean, 2000).

With the exception of MS, all FGFR-related craniosynostosis syndromes were originally defined phenotypically. Consequently, diagnosis of FGFR-related craniosynostosis syndromes is based on clinical findings (Robin et al., 1998) (e.g., craniosynostosis, dysmorphic facial features, limbs appearance) and, when possible, confirmed by genetic testing. This is important as genetic and phenotypic variation within and among craniosynostosis syndromes results in the now well-known lack of a one-toone correspondence between a given FGFR mutation and a specific skull shape. Some of the FGFR-related craniosynostosis syndromes can be caused by many different mutations within one or more of the FGFRs (e.g., PS and CS), while others are caused by a unique mutation (e.g., MS), or a very small set of mutations (e.g., AS) (Table 1). In addition, there are certain mutations in FGFR2 that have been associated with more than one clinical syndrome (e.g., CS and PS) (Passos-Bueno et al., 2008). Consequently, in many cases clinical diagnosis is problematic.

The craniofacial phenotypes of AS, CS, MS, PS are highly variable. Craniofacial phenotypes of AS can include varying degrees of midfacial retrusion. It is reported that AS cases carrying the FGFR2^S252W mutation have a more severe facial phenotype relative to AS patients who carry the FGFR2^P253R mutation while the FGFR2^P253R group has more severe limb anomalies (Slaney et al., 1996; Lajeunie et al., 1999; von Gernet et al., 2000). Craniofacial phenotypes of CS can vary from normal, to facial skeletal dys-morphologies without calvarial craniosynostosis, to cloverleaf skull malformation. In the majority of cases, several cranial sutures are prematurely fused at birth, although on occasion, the phenotypic features of CS may be absent at birth and evolve gradually during the first few years of life (Lajeunie et al., 1999; Connolly et al., 2004; Hoefkens et al., 2004). Variation in the severity of the craniofacial phenotype and limb anomalies of PS has led to the creation of three clinical subtypes (Cohen, 1993). Finally, the craniofacial phenotype of MS is characteristically variable and ranges from normal to severe (Doherty et al., 2007).

Facial phenotype is one of the key clinical findings used in differential diagnosis among the craniosynostosis syndromes. Although AS is characterized by a more dys-morphic facial skeleton relative to CS, MS, or PS (Cohen and MacLean, 2000), all of these syndromes share characteristic facial skeletal features (i.e., Crouzonoid face), including midfacial retrusion, hypertelorism, proptosis (secondary to orbital dysmorphogenesis), high-arched palate, flattened malar region, and beaked nose (Table 1) (Robin et al., 1998; Johnson and Wilkie, 2011). Midfacial retrusion or hypoplasia is defined as the posterior positioning and/or vertical shortening of the infraorbital and perialar regions, or increased concavity of the face and/or reduced nasolabial angle representing underdevelopment of the maxillary height (decreased midface height) or depth (retrusion of the maxilla) (Allanson et al., 2009). This definition allows for a wide range of variation and currently, midfacial retrusion represents a catch-all diagnosis for many craniofacial conditions that can result from distinct molecular causes and novel developmental dynamics. Midfacial retrusion is potentially the most challenging clinical manifestation of many of the FGFR-related craniosynostosis syndromes affecting oral health, feeding, and airway function (Cunningham et al., 2007; Johnson and Wilkie, 2011) but is relatively uncommon in MS (Ridgway et al., 2011).

Here, we perform a detailed three-dimensional (3D) evaluation of facial skeletal shape in a large retrospective sample (n = 43) of cases clinically and/or genetically diagnosed as AS, CS, MS, and PS to quantify variation in severity of facial dysmorphology, and precisely identify specific facial features pertaining to one or several of these four syndromes. First, we investigate the genotype phenotype correspondence for the segment of our sample (19/43) with known genetic mutations and subsequently add the remaining cases to further elucidate what knowledge of the causative FGFR mutation brings to our understanding of these syndromes.

The shape of the facial skeleton is a challenging clinical symptom and critical diagnostic feature that, along with craniosynostosis and limbs anomalies, is used to clinically diagnose FGFR-related craniosynostosis syndromes. Although based on a relatively small sample of genotyped patients (n = 19), our results reveal distinct facial phenotypes associated with FGFR2^S252W, FGFR2^P253R, and FGFR3^P250R mutations. Such robust genotype phenotype correspondence for AS and MS does not exist when the entire skull is analyzed. Our results show a quantitative scale of severity of facial dysmorphology and particularly midfacial retrusion that is strongest in cases carrying the FGFR2^S252W mutation, relatively decreased in the FGFR2^P253R group and further decreased in MS. The anatomical regions displaying the most intense shape differences among AS and MS when compared with unaffected individuals are the frontal and zygomatic contributions to the orbit, and the palate. Our results are in agreement with previous studies of humans with AS and mouse models for AS in which the facial skeleton, particularly the palate, was found to be the most profoundly dysmorphic cranial feature, especially in the FGFR2^S252W group (Slaney et al., 1996; Lajeunie et al., 1999; Von Gernet et al., 2000; Martínez-Abadías et al., 2010, 2013; Wang et al., 2010).

Three individual cases merit discussion. The CS case carrying the FGFR2^F276V mutation displayed no sign of craniosynostosis at the age of 2 months. Because of the absence of craniosynostosis, this case plots with the unaffected cases in the skull shape analysis (Supp. Fig. S5A), while in the facial skeletal shape analysis this individual plots with the other syndromic cases (Figs. 1A, 2A). The facial phenotype and the absence of characteristic limb anomalies prompted the diagnosis of CS for this individual, even though the FGFR2^F276V mutation has also been associated with PS (Passos-Bueno et al., 2008). These findings prompt us to propose that the FGFR2^F276V mutation functions to contribute to facial dysmorphology, perhaps by causing premature closure of facial sutures, but does not necessarily cause premature closure of the cranial vault sutures. The case carrying the FGFR2^A172F mutation that is specific to PS and known to be associated with severe limb phenotypes (Wilkie et al., 2002; Cohen, 2004), has a facial phenotype similar to that of AS (Figs. 1A, 2A). Finally, the FGFR2^C342R mutation is typically characterized by severe phenotypes (Lajeunie et al., 2006) and frequently associated with PS, although also with CS. Our case displays bilateral craniosynostosis of the coronal and lambdoid sutures and in our analysis is the second most severe facial phenotype after the PS case carrying the FGFR2^F276V mutation.

While genetic causes of CS include at least 47 distinct mutations mostly within FGFR2 (Passos-Bueno et al., 2008), and reduced dosage of ERF (Twigg et al., 2013), the quantitative range of facial shape variation for our total CS sample is small relative to AS (Fig. 2A), which is caused by only two mutations in 99% of cases. Obviously, a small fraction of the mutations causative for CS occurs at high incidence, and only a subset of these is most likely represented in our sample. Still, we can reasonably assume that some of the cases clinically diagnosed with CS carry a different mutation than the genetically diagnosed CS case included in the present study (i.e., FGFR2^F276V). Finally, the relatively small range of MS facial shape variation revealed by our analysis contrasts with current knowledge of variation in MS with phenotypic consequences ranging from normal to severe (Doherty et al., 2007). This most likely represents an ascertainment bias in that our sample is limited to individuals who sought treatment at a medical facility and would not include individuals with normal appearance or with a mild phenotype not requiring reconstructive surgery. Consequently, our MS sample most likely represents the more severe end of the MS phenotypic continuum.

Mutations causative for FGFR-related craniosynostosis syndromes do not always determine clearly distinguishable craniofacial phenotypes and variation is substantial. Our analyses show that, even if the mean shapes of the different syndromic groups are significantly different (as measured by the inter-group Procrustes distances), the within-group variation is large, particularly for skull shape, impeding separation between groups (as shown by the PCAs of the skull shape). It is worth noting that our precise, quantitative description of the similarities and differences between the 3D facial phenotypes of AS, MS, CS, and PS leads to groupings that are similar to those obtained when groups are defined on the basis of the affected functional domain of the specific FGFR (Table 1) (Cunningham et al., 2007). Indeed, AS, MS, and FGFR1-related PS are caused by mutations within the IgII-IgIII linker region, while the majority of CS and FGFR2-dependent PS are caused by mutations within the IgIII domain. The mechanistic effect of mutations within the IgII-IgIII linker region result in altered ligand-binding specificity and/or affinity (Anderson et al., 1998; Yu et al., 2000; Ibrahimi et al., 2001, 2004), while mutations within the IgIII domain result in aberrant intermolecular disulfide bonds between unpaired cysteine residues generating constitutive activation of the receptor (Neilson and Friesel, 1996; Mangasarian et al., 1997; Robertson et al., 1998). Although it is tempting to see a direct relationship between molecular data and our data on the corresponding craniofacial phenotypes, in the present study we identified two PS cases with facial shapes similar to AS and MS carrying FGFR2 mutations; the FGFR2^A172F mutation within the IgII domain and the FGFR2^C342R mutation within the IgIII domain, instead of the anticipated FGFR1 mutation within the IgII-IgIII linker region. Current knowledge suggests the importance of mechanisms of interaction among genes and modulation of gene expression by regulatory relationships in the making of clinical phenotypes. These dynamics, which are becoming clearer for some mutations (Sharma et al., 2013) will most likely provide the mechanistic explanation for levels of penetrance observed in patients carrying craniosynostosis mutations.

Facial reconstructive surgery, necessary in the more severe craniosynostosis cases is aimed at improving function and cosmetic appearance. Our analysis of facial skeletal dysmorphology in craniosynostosis syndromes defines distinct phenotypes and patterns of variation for diagnostic groups based on genetic and clinical information, and this should be no surprise. What our analysis brings to the field is quantitative evidence that facial dysmorphogenesis within these syndromes is not a generic event. Rather, each case of craniosynostosis results from change(s) in specific (and potentially diverse) signaling and regulatory cascades, and these changes have multiple consequences for developmental trajectories and phenotypic variation. The overall similarity in facial morphology among craniosynostosis syndromes (i.e., the Crouzonoid face) underscores the fact that generally similar phenotypes can be produced by different mutations in genes within the same or similar gene expression cascades. The differences in intensity of shared dysmorphologies observed in facial phenotypes of AS, MS, PS, and CS (Figs. 1, 2) may, with further data, provide key information about exactly which developmental pathways are being activated in a specific individual. Ideally, the synthesis of quantitative phenotypes, precise knowledge of the molecular consequences of specific mutations and their impact on developmental dynamics will reveal the mechanisms underlying craniosynostosis phenotypes providing valuable information to clinicians and surgeons managing these patients. The critical upshot would be to predict precise, individual phenotypes given genetic information (Houle, 2010).

Unraveling the complex relationship between genotype and phenotype requires levels of precision in analysis of phenotypes comparable to those used in modern genetic analysis. Our suggestion is that equally precise phenotypic analyses should accompany next-generation genetic research and that these approaches should proceed cooperatively, both in the study of animal models and humans. Not only will surgical planning be improved but new opportunities and limitations for the development of therapeutic interventions will be revealed.