Three‐dimensional facial morphology in Cantú syndrome

Abstract Cantú syndrome (CS) was first described in 1982, and is caused by pathogenic variants in ABCC9 and KCNJ8 encoding regulatory and pore forming subunits of ATP‐sensitive potassium (KATP) channels, respectively. It is characterized by congenital hypertrichosis, osteochondrodysplasia, extensive cardiovascular abnormalities and distinctive facial anomalies including a broad nasal bridge, long philtrum, epicanthal folds, and prominent lips. Many genetic syndromes, such as CS, involve facial anomalies that serve as a significant clue in the initial identification of the respective disorder before clinical or molecular diagnosis are undertaken. However, an overwhelming number of CS patients receive misdiagnoses based on an evaluation of coarse facial features. By analyzing three‐dimensional images of CS faces, we quantified facial dysmorphology in a cohort of both male and female CS patients with confirmed ABCC9 variants. Morphometric analysis of different regions of the face revealed gender‐specific significant differences in face shape. Moreover, we show that 3D facial photographs can distinguish between CS and other genetic disorders with specific facial dysmorphologies that have been mistaken for CS‐associated anomalies in the past, hence assisting in an earlier clinical and molecular diagnosis. This optimizes genetic counseling and reduces stress for patients and parents by avoiding unnecessary misdiagnosis.

The underlying reason for this heterogeneous clinical presentation is still unclear.
Dysmorphic facial features, on the other hand, are observed in every CS patient and are usually evident at birth. In younger patients, features include a low frontal hairline, epicanthal folds, flat nasal bridge, anteverted nares, long philtrum, macroglossia, prominent mouth, and thick lips. Additionally, the palate may be high arched and/or narrow, the gingiva may be thickened and an anterior open bite may be present . With advancing age, however, these clinical features are subject to considerable phenotypic changes. Whereas the initial facial shape appears round with full cheeks, the facial phenotype progresses to a lengthening of the face with a pointed chin and a more prominent forehead (Scurr et al., 2011). Progressive coarsening of the face results in further flattening of the nasal bridge with prominent supra-orbital ridges and fuller lips.
The small nose becomes more bulbous. Periorbital fullness seems to be consistent (Roessler et al., 2018).
Despite the identification of ABCC9 and KCNJ8 pathogenic variants as being causal, an explanation for variability of CS-associated features remains unknown. The correlation between GoF mutations in K ATP channels and the coarse facial appearance is currently still not understood. Additionally, in a recent report describing 74 patients in the International Cantú Syndrome Registry (ICSR) we were unable to correlate genotype to specific phenotypic features in CS individuals including facial appearance . features with mucopolysacharidosis being most common; an unnecessary and stressful process for parents . Hence, quantitative analysis of the described facial morphology is of vital importance in characterizing the phenotype, which will assist in achieving earlier clinical and molecular diagnosis and thus optimizing appropriate medical care and genetic counseling.
In the present study, we investigated the effect of CS-associated mutations in ABCC9 on craniofacial structures using DSM based analysis of 3D photographs of 20 clinically affected and molecularly proven CS patients. Additionally, we compared facial dysmorphism in CS with genetic disorders that can be a differential diagnosis of CS because of overlapping facial features. Our study assembles the second largest cohort of CS patients studied so far, considering the overall rareness of the disorder .
We demonstrate that DSM-based analysis provides a very accurate instrument to classify faces or facial regions along the CS-control spectrum. In addition, we highlight specific facial differences between CS and genetic disorders with similar dysmorphisms to improve initial diagnosis of these patients. As a result, our findings demonstrate the utility of 3D imaging during a genetic diagnostic process.

| Study participants
All data reported here were collected after informed written consent of the patient or the parents or legal guardians for patients younger than 18 years old.
The CS patients group comprised 10 male patients 2.7-16.2 years old and 10 female patients 3.1-22.0 years old. Figure S1 shows distribution of country of origin over age in CS patients. All patients had been molecularly diagnosed with CS either by Sanger-sequencing of the ABCC9 gene or whole exome sequencing. All study participants are part of the ICSR and have been reported previously at least once . 3D face images of individuals with CS were collected during special annual Cantú syndrome research clinics at the Utrecht University Medical Center, the Netherlands, and at Washington University, St. Louis, MO. The accuracy of such 3D imaging devices has been shown to be highly reliable (Aldridge, Boyadjiev, Capone, DeLeon, & Richtsmeier, 2005;Camison et al., 2018). Some images were unusable because of incomplete face coverage or poor subject cooperation. Details of the specific ABCC9 variants and body weight and height were obtained from the ICSR .
Images of healthy subjects as well as patients diagnosed with Williams syndrome (WS), Fabry disease (FD), Rubinstein-Taybi syndrome (RTS) and Rasopathies were selected from an existing collection.
All individuals used in this study are Caucasian but from different countries of origin. Figure S2 shows distribution of country of origin of controls and patients with another disorder used in this study.

| 3D image capture and preparation
A total of 20 3D face images of individuals with CS were captured with commercial photogrammetric devices (3dMDface System [3dMD Inc.] and Canfield Vectra 3D system [Canfield Scientific Inc.]). All images were manually annotated by a single operator (PH) with 22 facial landmarks previously shown to be accurate and reproducible (Toma, Zhurov, Playle, Ong, & Richmond, 2009).

| DSM building and closest mean classification
A dense surface model of a set of landmarked face surfaces, described in detail elsewhere (Hammond, 2007;Hutton, Buxton, Hammond, & Potts, 2003), comprises shape variation modes arising from a principal component analysis (PCA) of differences of the surface points' positions from those of a comparator average face. Prior to the PCA, using a base mesh (whole face or patch) and aligned sparse landmarks, a dense correspondence of surface points across all faces is induced with no manual interaction. The proportion of face shape variance covered by each DSM mode is computed, and typically the modes are ordered in terms of decreasing variance coverage. For a DSM of a collection of faces of children and adults, the first mode typically reflects facial growth and correlates strongly with age (Hammond, 2007). We retained sufficient modes to cover 99% of shape variance in each constructed DSM.
Each face surface captured has as many as 30,000 mesh points.
The signature of a face surface is the set of position differences at constituent image mesh points from corresponding points on the mean of age/sex-matched healthy controls, normalized against the variation in controls. The signature weight of a surface is the square root of the sum of the squared normalized differences across all of the densely corresponded points. Signature weight is a rough estimate of facial dysmorphism. A signature heat map visualizes the significance of localized differences using a red-green-blue spectrum with, for example, red and blue reflecting extreme opposite displacements and green coincidence with the mean of the matched controls. Thus, an axis normal to the face surface reflects inward/outward displacement.
In order to investigate differences in face shape, we used multifolded discrimination testing (closest mean; support vector machines; linear discriminant analysis) to determine baseline discrimination rates between controls and syndromes. Anthropometric comparisons were made against appropriate age-range matched controls.

| Comparison of linear regressions
Linear regressions were undertaken for various facial measurements and DSM-based markers against age. Prism (Graphpad) was used to determine significant differences in slope and/or intercept in comparisons of separate regressions for control, CS and other disorder cohorts.

| Statistical analysis
Significance values when comparing gender-specific facial features of CS patients and controls were calculated using an unpaired Student's t-test throughout the article. Statistical robustness was further enhanced through applying a Bonferroni correction that divides the pvalue cut-off by the number of tests run. To find statistically significant differences in facial features between male and female CS patients and age-and gender-matched controls, 5% p-value cut-off was divided by 4, resulting in a threshold p-value of .0125.

| RESULTS
We present 3D imaging analysis in 20 CS patients. All patients have among patients was similar to those reported in previous studies.

| Patient demographics
The 20 patients (10 females and 10 males) range in age from 2.7 to 22.0 years (mean: 11.46 years) ( Comparison of the mean faces of the CS and control group confirms many of the previously published facial features. In Figure 1

| Facial growth and overall dysmorphism in CS
In a face DSM with mixed age range, the first principal component When calculating PC1 for all four age-and gender-specific subgroups facial growth is still significantly increased in both male subgroups (age range 2-5 years: p = .0004, age range 10-22 years: p = .001), whereas facial growth seems to become more similar to healthy controls in female CS patients (age range 2-5 years: p = .0713, age range 10-22 years: p = .0776) ( Figure S8a,b).
The overall dysmorphism score was calculated as the square root of the sum of squared differences of displacements in the signature heat map comparison with the mean of age-and sexmatched controls. Firstly, this score suggests that facial dysmorphism does not increase much over time in CS patients ( Figure 4b). However, a comparison between female and male patients shows significantly more dysmorphism across all ages in male patients (p = .0151). Calculation of signature weight in all four age-and gender-specific subgroups confirms these observations ( Figure S8c).  Face size and growth in CS patients, interpreted by PC1 in the model, is always greater compared to age-matched controls. Overall facial dysmorphism, is significantly higher in male CS patients compared to females but unlike the other syndromes investigated, it barely increases or even stagnates with age in all studied subgroups. Notably, we would like to point out that all analysis performed in the four ageand gender-specific CS subgroups can merely be seen as an indication of facial development in CS since the subject numbers are small (3-7 patients) and therefore miss statistical robustness. Thus, we strongly suggest to primarily focus on data aquired from CS subgroups containing all male or female patients as shown in Figures 2 and 3. In order to properly assess changes in facial features in different CS age groups one would require a cohort with significantly more subjects.
Our results are mostly in line with the features described in various clinical reports that assessed facial characteristics of CS patients based on medical photography (Roessler et al., 2018;Scurr et al., 2011). However, increase in mouth width which has been previously reported in the literature (Grange, Nichols, & Singh, 2014;Roessler et al., 2018) was not observed applying 3D image analysis. Hence, we demonstrate that 3D imaging recognizes common facial anomalies observed in CS patients. Since the results generated in this study are produced by objective quantitative analysis and are thus not hampered by inter-observer variability, the applied approach enables a more reliable assessment and quantification of facial anomalies observed in CS patients. Moreover, we include comparison of both male and female CS patients with respective controls as well as evaluation of facial growth and overall dysmorphism.
Notably, our study cohort only includes Caucasians, originating from both the Netherlands and the United States with an age range of 2.7-22.0 years, which resembles the overall patient demographic reported in literature so far . Even though all applied control subjects in this study are also Caucasian, they originate from two European populations of close phylogenetic and geographic proximity, the UK and The Netherlands ( Figure S2). Even though previous studies have shown facial differences in different adult Caucasian populations (Hopman, Merks, Suttie, Hennekam, & Hammond, 2014), we do not expect these to be present in younger individuals.
The uneven age distribution of patients, which can be observed in our CS cohort, might be seen as a limitation when making simple anthropometric comparisons with controls but for shape specific considerations like face signature each individual is normalized against a sex-and-age matched set of controls which is more reliable.
It would be of interest to perform a similar study in CS patients with an increased age as well as from non-European ancestry, which would require a larger cohort as is identified currently.  . The majority of misdiagnoses were based on facial dysmorphology and hypertrichosis and could however not be confirmed by additional genetic of metabolic testing.
Since these misdiagnoses implicate a completely different prognosis and life expectancy, earlier CS face recognition could prevent unnecessary stress for parents.
By comparing facial characteristics of CS with patient cohorts whose facial anomalies have been mistaken with CS in the past we demonstrate that 3D imaging analysis can be successfully applied to delineate between syndromes and thus avoid further misdiagnoses. It would be of interest to perform a similar comparison between CS and MPS patients in the future. Notably, congenital hypertrichosis, also observed in all CS patients , can sometimes be applied as clear discriminator between CS and other mentioned syndromes, if severe enough.
Most importantly, the described 3D imaging technique reveals multiple advantages including practicality, availability and low costs.
3D cameras and appropriate computer software for analysis are available in every clinic that makes assessing facial morphology via 3D imaging quick and cheap. Hence, the implementation of this in general practice for clinical diagnosis of CS patients but also other disorders with a facial component is feasible.
Recent innovations based on 2D technologies combined with computer vision and deep learning algorithms also hold promise to be worthwhile in diagnosis of genetic disorders with facial components (Gurovich et al., 2019;van der Donk et al., 2019). Although 2D images are more readily available, they are subject to greater variation in quality and pose due to projection distortion, and more sophisticated analysis is required to extract the discriminating features. Thus, a significantly increased number of images is required to train the pattern recognition algorithms. Therefore, 2D technologies may represent a considerable alternative in clinics to successfully diagnose CS and other genetic disorders with facial anomalies in the future, but it still remains to be seen which approach has more potential to be established in a stable fashion.

| CONCLUSION
In conclusion, we report a valid method for quantification of facial dysmorphic features in CS.
Although facial dysmorphism in CS can be clinically recognizable, an objective, quantitative evaluation is especially valuable when assessing phenotypically or genotypically unusual cases.
The genes associated with CS have only recently been identified.
Hence, we suspect there may be more affected individuals-particularly older patients-who have not received genetic testing and therefore remain undiagnosed. We demonstrate that 3D imaging analysis can be confidently applied to unravel characteristic facial features of CS and successfully discriminate between the syndrome and other genetic disorders it has been confused with in the past due to facial similarities.
This will help in diminishing parental anxiety related to misdiagnoses.
In the future, 3D-based analysis will enable the recognition of CS facial characteristics in patients who have not had an adequate clinical diagnosis. Subsequent molecular analysis of such patients may identify mutations and in combination with detailed facial analysis enable unequivocal identification of causative genes. Considering the current efforts to develop a pharmacological treatment for CS (Ma et al., 2019;McClenaghan et al., 2019), 3D technology additionally may enable the evaluation and quantitative measurement of effective response to therapy options in the future which will possibly lead to a decrease in coarseness of the face.