Autism is a spectrum of disorders (ASDs) united by a common Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV)-defined [1] behavioral phenotype. Research into this disorder is increasingly focused on both genetic causes and neuroanatomical bases for the behavioral phenotypes. Thus far attempts to discover major autism susceptibility genes have been largely unsuccessful, with approximately only 15% to 20% of cases of autism linked to specific genes, chromosomal aneuploidy or recognized syndromes [2–4]. The rest remain idiopathic. A primary reason for the lack of progress in understanding the etiology and genetic underpinnings of ASD is undoubtedly the significant heterogeneity within both behavioral and clinical phenotypes.

The neurodevelopmental model of ASD [5] suggests that changes in embryonic developmental patterns result in the spectrum of ASD phenotypes and that these changes may result from permutations of genetics, the environment, or the interaction of the two. This model posits that the brain is altered during embryonic development, a time when the brain is intimately tied to developing facial tissues via genetic signaling, biomechanical and biochemical mechanisms [6–13]. The face and brain form a vast but segmented population of cells whose differentiation and identity are established through intricate signaling mechanisms. It has been stated that the brain is the foundation on which the various parts of the developing face grow [14]. The phrase, "The face predicts the brain" [15], has been employed frequently to explain developmental disorders such as holoprosencephaly. Thus changes to the developing brain may be reflected in the face [15–17].

The face develops from populations of neural crest cells migrating from the neural tube into developing embryonic facial prominences. These neural crest cells interact with the developing brain via both physical contact and genetic signaling. Previous research has shown that the expression of Sonic hedgehog (SHH), fibroblast growth factor 8 (FGF8) and bone morphogenetic proteins (BMPs outline developmental interactions between the face and the brain in discrete temporal and regional patterns [18–22]. The connection of brain and facial phenotypes to their underlying genetic bases involves hierarchies of complex regulatory cascades, nested epigenetic networks and ever-changing patterns of cross-talk between molecules, cells and tissues throughout development [6, 7, 12, 23]. Localization of facial phenotypic variations to specific areas of the face may reveal potential candidate genes and/or pathways targeted in the development of the brain in autism.

Given the clear evidence that the embryological face and anterior brain emerge and develop in exquisite intimacy [16, 17], facial phenotypes can serve as accessible and informative indices of brain phenotypes in neurodevelopmental disorders. We propose that altered expression of genes involved in the development of neural tube structures and overlying facial prominences may result in distinct facial and neural phenotypes in ASD. We hypothesize that there are common autism-causing genes that affect early brain development and simultaneously the facial phenotype. Defining one or more common facial phenotypes within ASD will provide a new physical biomarker that can be used to improve ASD diagnoses, with all the associated benefits related to prognosis, recurrence counseling, choice of subgroup-appropriate therapies and the possibility of developing a screening tool to assist in early diagnosis.

Precise measures of phenotypes and innovative methods of analysis are integral to discovering the nature of developmental contributions to phenotypic variation. Using state-of-the-art three-dimensional photography to obtain facial images to precisely measure facial phenotypes, we tested two hypotheses. First, children with ASD display a subtle but distinct facial phenotype compared to typically developing (TD) children. Second, there are subgroups of facial phenotypes within the population of children with ASD that denote biologically discrete subgroups.

Previous work has suggested that there are autism facial phenotypes with a developmental basis. In an epidemiological study of facial photographs of children with autism and developmental disabilities, Rodier and colleagues [24, 25] reported a facial phenotype common in autism consisting of decreased interpupillary distance (although intercanthic distance was not decreased), ptosis, strabismus, lop ears and hypotonia of the lower face. They postulated that these minor anomalies arise as the face is closing and the cranial nerves are invading mesenchyme that will develop into the muscle, skeletal and dermal tissues of the head. Rodier and colleagues [26, 27] also suggested that facial phenotype might allow researchers to pick out the children whose autism is due to mutations in the homeobox genes which control the development of both the brainstem and face. Though this group of genes is important for embryological development, we now know that the face and the rostral brain are not patterned by genes in the homeobox family [28]. Additionally, Hammond et al. [29] studied a group of boys, ages 2 to 18 years from families with at least two affected family members, finding minor shape differences in comparisons of the mean facial phenotype of these boys to a control mean facial phenotype, with their major findings emphasizing significant facial asymmetry in boys with ASD and their family members and suggesting a shared developmental basis for these phenotypes.

The embryonic face is derived from seven prominences that come together to form a face. These include the midline frontonasal process (FNP) and the paired lateral nasal prominences (LNPs) as well as maxillary prominences (MAXs) and mandibular prominences (MANDs) (Figure 1). The LNPs are very quickly assimilated into the FNP. Over the course of embryonic development, the facial mesenchyme of these developing prominences is bounded by the epithelia derived from both the forebrain neuroectoderm and the facial ectoderm [18]. The midline FNP forms from neural crest cell populations arising on the surface of the forebrain, migrating over the forebrain to become encased within the neural ectoderm of the forebrain and the facial ectoderm [19]. In fact, signals from the forebrain neuroectoderm are essential for the survival of the neural crest cells of the FNP [30], including SHH, FGF8 and BMP2 signaling [18–22]. This prominence gives rise to the forehead, the midline of the nose and the oral philtrum (Figure 1). LNP, MAX and MAND are also formed from neural crest and mesoderm cells, as well as from the epithelia of the facial surface ectoderm and pharyngeal endoderm [31]. These three laterally developing prominences are highly responsive to Wingless-type (WNT) signaling, whereas the midline FNP is not [32]. Similarly, regionalization of the neural tube is controlled by the SHH, FGF and WNT families [20]. SHH is expressed in the facial ectoderm and neuroectoderm at various developmental stages [31]. Thus face and brain in the embryo develop in concert, both temporally and genetically, and altered phenotypes of the face should reflect altered phenotypes of the brain via their shared developmental program.

We proposed that ASD might be associated with a distinctive facial phenotype and hypothesized that we might be able to identify subgroups within ASD based on facial phenotypes. Using sophisticated facial phenotyping based on three-dimensional stereophotogrammetric imaging and advanced statistical analyses, we developed a research methodology that allowed us to test and garner support for both hypotheses. In addition, we maximized our chances of success by selecting a relatively homogeneous ASD population that included only boys of self-reported Caucasian ethnicity, with essential autism, defined as having no discernible dysmorphology or microcephaly and within a limited age range.

We found that essential autism in boys is associated with a distinctive facial phenotype characterized by an increased breadth of the mouth, orbits and upper face, combined with a flattened nasal bridge and reduced height of the philtrum and maxillary region. This facial phenotype is similar to the one we recognized clinically (JHM) and may be the "beautiful face" mentioned by Kanner [58]. Embryologically, this facial phenotype is indicative of a perturbation of the FNP. It can be explained by reduction in the superoinferior dimension of the midline structures derived from the FNP, an increase in the subnasal portions of the FNP and a concomitant increase in the breadth of the upper face.

The common facial phenotype described by Rodier and colleagues [25] includes a reduced interpupillary distance with no difference in intercanthic distance. In contrast, in our present study, we found a narrowing of the intercanthic distance, or mild hypotelorism. These findings are complementary in that we found an overall decrease in intercanthic distance, which potentially translates to decreased interpupillary distance, although we did not directly measure that distance. Our findings are also in line with those of Hammond et al. [29]. However, the sample included in the Hammond et al. [29] study consisted of a group of boys, ages 2 to 18 years from families with at least two affected family members. Our study extends the findings of both Rodier and colleagues [25] and Hammond et al. [29] by quantifying precisely localized differences and variations in facial phenotypes in a homogeneous group of boys with essential autism.

Our findings further extend previous work in that we have discovered two subgroups of boys with ASD who displayed unique facial phenotypes, which correspond to distinct clinical phenotypes, compared to both the majority of boys with ASD and TD boys. Subgroup 1 displays decreased height of the facial midline and increased breadth of the mouth as well as the length and height of the chin. These regions of the face develop primarily from the FNP and midline portions of the MAND prominences of the embryonic face. Subgroup 2 displays increased breadth of the upper face in combination with decreased height of the philtrum. Both of these regions develop from the embryonic FNP.

The results of the tests of our hypotheses indicate that boys with ASD have an altered developmental pattern of the structures derived from the embryonic FNP and the MAND. It is well-documented that the developing FNP is derived from localized, specific cell populations under patterned genetic control. A number of developmental genes have been implicated in patterning the outgrowth of FNP, including FGF8, SHH and BMP2.

Neural crest cells that ultimately make up the FNP migrate over the forebrain to become encased within the neural ectoderm of the forebrain and the facial ectoderm [19]. FGF8 plays a chemoattractive role in neural crest cell migration [20]. Signals from the forebrain neuroectoderm are essential for the survival of the neural crest cells of the FNP [30], including SHH-dependent signaling from forebrain [21]. Other studies have shown that SHH provides a key signal in regulating facial neural crest cell survival and patterning [22] and regulates BMP2 expression in the middle and upper face [18]. Furthermore, it has been shown in an animal model that decreased SHH signaling leads to narrowing of the FNP and hypotelorism [59]. The MAXs and LNPs are responsive to WNT signaling [32]. WNT signaling leads to elevated cell proliferation and migration of neural crest cells [32]. No midline structures are responsive to WNT.

Correct patterning and development of the forebrain also requires a balance of these genetic signaling factors [18], including SHH and the FGF, BMP and WNT families [20, 60]. SHH is expressed in the facial ectoderm and neuroectoderm at various developmental stages and acts synergistically with FGF8 in both face and brain development [31]. SHH expression in facial ectoderm affects the expression of FGF8 in the brain [30]. Likewise, expression of FGF8 by the forebrain is stimulated by the presence of neural crest cells [61].

It is clear that development of the face and brain is an interactive process, both anatomically and genetically. Gene expression studies have shown that facial and neural tube development are intimately interrelated. Altered gene expression patterns are associated with alterations in face and brain development. The reverse is also true: Altered face and brain development is associated with alterations in gene expression. We know that the brain is altered in people with ASD (reviewed in [62]), and the results of our present study show that the face is also affected. However, the sequence of events leading to these differences is unclear and may differ among the various subgroups described herein.

There is evidence derived from genetic studies implicating the developmental genes that control the patterning of the FNP and forebrain which causes autism. The SHH gene, though not identified by autism linkage or association studies, is functionally related to the Patched gene (PTCHD1), which is a strong autism candidate gene. Investigators who conducted CNV studies [63] first identified PTCHD1 gene microdeletions and missense mutations in males with ASD [64]. SHH signaling repression is relieved when SHH binds to PTCHD1. BMP is part of the functional face and brain patterning network interacting with SHH and FGF8 to maintain brain and face patterning. Bakrania et al. [65] evaluated gene expression in embryos and demonstrated cotemporal and cospatial expression of BMP4 and SHH signaling genes. It is expected that sophisticated brain and/or face functional transcriptome studies may be useful in further linking genes involved in simultaneous face and brain development.

The clinical and behavioral differences that we identified between subgroup 1 and subgroup 2 boys support our hypothesis that the subgroups are biologically and etiologically distinctive. Subgroup 1 appears to be more severely autistic, with only 8% diagnosed with Asperger syndrome compared to 60% in subgroup 2, and has higher SCQ lifetime scores than subgroup 2 (25.3 vs 17.7). The percentage of IQ scores less than 70 were higher in subgroup 1 than in the other groups: 50% on the FSIQ and 45% on the NVIQ, compared to only 20% on both the FSIQ and NVIQ (one of five) in subgroup 2. and 15% FSIQ and 12% NVIQ in the ASD remainder group. Verbal IQ scores were also lower in subgroup 1 but did not reach statistical significance. Subgroup 1 also displayed several features predictive of poor outcome, including a higher risk for seizures and increased incidence of language regression at ASD onset. The observation of significantly higher SRS scores in mothers of boys in subgroup 1 is interesting but unexplained at this time.

By contrast, subgroup 2 appears to be aligned more with an Asperger syndrome diagnosis, which was made in 60% of subgroup 2 participants compared to only 8% in subgroup 1 and 32% in the remainder. Lifetime SCQ scores in subgroup 2 were also lower (17.7) than those in subgroup 1 (25.9) and the remainder (20.6). Consistent with the Asperger syndrome diagnosis, boys in subgroup 2 spoke their first words significantly earlier than boys in subgroup 1 and the remainder, and they were significantly more likely to be macrocephalic (80% vs 17%) compared to subgroup 1. Though IQ score differences did not reach significance within this small subgroup, boys in subgroup 2 had consistently higher FSIQ, NVIQ and VIQ scores than boys in subgroup 1.

Though we maximized our study outcomes by using a relatively homogeneous group of boys of Caucasian ancestry with essential autism within a narrow age range, the subject group was imperfect in a number of ways. Though 56% of the boys had participated in the SSC, not every participant completed the entire test battery, which slightly decreased the number of subjects who could be analyzed statistically for some comparisons. However, all of the four boys without VIQ scores and the one boy without a FSIQ score clustered morphologically within the main group and not within either subgroup. Thus our findings are highly unlikely to be affected by the small number of missing data points. In addition, the SSC population is biased toward a higher-functioning group of boys, which tended to shift the cognitive and outcome curves. The higher-functioning population recruited from the SSC may, however, have helped delineate subgroup 2 by increasing the proportion of subjects with Asperger syndrome. In addition, the number of boys in each subgroup was small. One additional observation is that in subgroup 2, which contained only five boys, one of the five was an outlier with significantly lower scores on IQ, language and outcome measures. This suggests that in future studies of larger numbers of subjects, we may find that subgroup 2 will be dissected into subgroups 2A and 2B. Finally, although one of the strengths of our study is that it comprised a homogeneous group of boys, that is, narrow age range, single sex, limited ethnic diversity and diagnosis made by a single clinician, it remains to be seen whether our findings will be consistent in a more heterogeneous population.

Differences in facial morphology may reflect alterations in embryologic brain development in children with ASD. Our results suggest potential differences in etiologies for the various subgroups of children. Further investigations into brain morphology will help to elucidate the causes and significance of these subtle differences. Verification of the role of a number of neurodevelopmental candidate genes may also be expedited by restricting analyses to the more homogeneous autism subgroups described herein. Likewise, based on our understanding of facial and neural development, identification of specific neurodevelopmental genes responsible for autism suggests which regions of the embryonic brain are most apt to be affected, providing potential target structures for future investigation.

Written informed consent was obtained from the parents of the patients for publication of this research article and any accompanying images. A copy of the written consent is available for review by the Editor-in-Chief of this journal.