Association Between Prenatal Alcohol Exposure and Craniofacial Shape of Children at 12 Months of Age

Key Points

Question  Is there an association between different levels of prenatal alcohol exposure and child craniofacial shape at 12 months?

Findings  This cohort study conducted an objective and sensitive craniofacial phenotype analysis of 415 children, which showed an association between prenatal alcohol exposure and craniofacial shape at almost every level of exposure examined. Differences in the midface and nose resemble midface anomalies associated with fetal alcohol spectrum disorder.

Meaning  Any alcohol consumption has consequences on craniofacial development, supporting advice that complete abstinence from alcohol while pregnant is the safest option; it remains unclear whether the facial differences are associated neurocognitive outcomes of prenatal alcohol exposure.

Importance  Children who receive a diagnosis of fetal alcohol spectrum disorder may have a characteristic facial appearance in addition to neurodevelopmental impairment. It is not well understood whether there is a gradient of facial characteristics of children who did not receive a diagnosis of fetal alcohol spectrum disorder but who were exposed to a range of common drinking patterns during pregnancy.

Objective  To examine the association between dose, frequency, and timing of prenatal alcohol exposure and craniofacial phenotype in 12-month-old children.

Design, Setting, and Participants  A prospective cohort study was performed from January 1, 2011, to December 30, 2014, among mothers recruited in the first trimester of pregnancy from low-risk, public maternity clinics in metropolitan Melbourne, Australia. A total of 415 white children were included in this analysis of 3-dimensional craniofacial images taken at 12 months of age. Analysis was performed with objective, holistic craniofacial phenotyping using dense surface models of the face and head. Partial least square regression models included covariates known to affect craniofacial shape.

Exposures  Low, moderate to high, or binge-level alcohol exposure in the first trimester or throughout pregnancy.

Main Outcomes and Measures  Anatomical differences in global and regional craniofacial shape between children of women who abstained from alcohol during pregnancy and children with varying levels of prenatal alcohol exposure.

Results  Of the 415 children in the study (195 girls and 220 boys; mean [SD] age, 363.0 [8.3] days), a consistent association between craniofacial shape and prenatal alcohol exposure was observed at almost any level regardless of whether exposure occurred only in the first trimester or throughout pregnancy. Regions of difference were concentrated around the midface, nose, lips, and eyes. Directional visualization showed that these differences corresponded to general recession of the midface and superior displacement of the nose, especially the tip of the nose, indicating shortening of the nose and upturning of the nose tip. Differences were most pronounced between groups with no exposure and groups with low exposure in the first trimester (forehead), moderate to high exposure in the first trimester (eyes, midface, chin, and parietal region), and binge-level exposure in the first trimester (chin).

Conclusions and Relevance  Prenatal alcohol exposure, even at low levels, can influence craniofacial development. Although the clinical significance of these findings is yet to be determined, they support the conclusion that for women who are or may become pregnant, avoiding alcohol is the safest option.

Prenatal alcohol exposure (PAE) is a major preventable cause of health and developmental problems in children. It may cause irreversible damage to the brain, resulting in fetal alcohol spectrum disorder (FASD), which is characterized by learning difficulties, executive dysfunction, impaired speech, motor problems, and behavior problems.¹ Fetal alcohol spectrum disorder may affect 3% to 5% of mainstream school-aged children, with many remaining undiagnosed.² Fetal alcohol syndrome (FAS) is diagnosed when cognitive impairment occurs together with abnormalities of growth and a characteristic facial phenotype. Diagnostic criteria of FAS include short palpebral fissures, flat philtrum, and thin upper lip,³ but the condition has also been associated with midfacial and auricular anomalies.⁴ Detailed morphologic studies of the face have found a reduction in ear length and midfacial hypoplasia, evidenced by reduced midfacial depth and flattening of the nasal bridge and malar regions and a reduction in the size of the neurocranium and face (reduced upper facial width, head circumference, and total facial height), as well as retrognathia and micrognathia (reduced bigonial breadth), in individuals who received a diagnosis of partial FAS and in individuals with heavy PAE who do not meet the diagnostic criteria for FASD.^5-7 Furthermore, comparisons of the physical characteristics of children with FASD and the physical characteristics of children without FASD have shown that maternal drinking measures significantly correlated with facial dysmorphology, with higher levels of drinking predicting higher dysmorphology scores.^8-10 Together these findings suggest a possible dose-related association between PAE and craniofacial shape.

Most of the studies cited have used facial measurements, which capture limited information and are prone to measurement error, or used clinical examination, which may be subjective. In 2013, Suttie and colleagues⁵ used objective, spatially dense morphometric techniques to analyze 3-dimensional (3-D) photographs in a social setting in which high alcohol intake is common. These techniques enable the analysis of the whole facial surface and do not require subjective assessment. In their study, FAS and partial FAS phenotypes were characterized by the difference between the average facial shape of control individuals and individuals with FAS or partial FAS. They also performed a clustering analysis of individual faces and found that some children with heavy PAE displayed anomalies consistent with FAS or partial FAS. Here, we extend this work to examine the phenotype of children who did not receive a diagnosis of FASD but who had lower levels of PAE, by using a shape regression–based approach, which allows us to control for covariates while comparing average faces statistically.¹¹

Asking Questions About Alcohol in Pregnancy (AQUA) is a population-based longitudinal study of the neurodevelopmental outcomes in children with PAE, with a focus on low to moderate alcohol consumption.¹² Participants are 1570 women and their offspring, recruited in early pregnancy during a 12-month period from July 25, 2011, to July 30, 2012, from low-risk public maternity clinics in Melbourne, Australia. Three-dimensional craniofacial images were captured at 12 months from a subset of 517 participants. Included in this analysis are the images of 415 white children whose mothers were not lifetime abstainers and for whom we have complete information on all covariates. The AQUA study was approved by the Human Research Ethics Committees of Eastern Health (E54/1011), Mercy Health (R11/14), Monash Health (11071B), the Royal Women’s Hospital (11/20), and the Royal Children’s Hospital (31055A) in Melbourne, Australia. Mothers provided written consent prior to image capture.

Questionnaires were collected that had detailed information on the quantity and frequency of alcohol consumption for the 3 months before pregnancy and for each trimester, including the time prior to recognition of pregnancy. Frequency of drinking and typical amount and type of alcoholic drink were combined to provide a single-exposure measure for each stage of pregnancy, expressed in grams of absolute alcohol (AA) and using algorithms previously described.^13-15

Exposure levels were low (≤20 g of AA per occasion and ≤70 g of AA per week), moderate (21-49 g of AA per occasion and ≤70 g of AA per week), high (>70 g of AA per week), and binge (≥50 g of AA per occasion).¹² Mothers who were abstinent throughout pregnancy comprised the control group for all analyses.

Analysis used a 3-tiered approach in which PAE tier 1 consisted of children of women who drank any alcohol while pregnant. Tier 2 subdivided the exposure group into those with PAE in the first trimester only and those with PAE throughout pregnancy. Tier 3 further subdivided the exposure group into low, moderate to high, or binge-level drinking before becoming aware of pregnancy and whether exposure occurred in the first trimester only or throughout pregnancy.

Medical photographers not involved in the analysis collected 3-D craniofacial images between January 8, 2013, and February 11, 2014, inclusive, within 2 weeks of the child’s first birthday, at the Royal Children’s Hospital in Melbourne, Australia. Image acquisition was via the 3dMD 7-pod system (3dMD Corp), which captures a 360° image of the head, including the face and cranium. To ensure that images were unobscured by hair and to capture the shape of the neurocranium, a tight-fitting stocking was placed over the cranial vault. A neutral facial expression was not ensured at all times, and some images were taken with the mouth open and others with the mouth closed. Mouth position in the images was subsequently standardized individually using a previously published method¹⁶ to ensure that each image had a neutral expression.

Craniofacial measurement was undertaken by a researcher (H.M.) blinded to the participants’ PAE group. To represent the entire surface of the face, a spatially dense array of 69 587 points on a template 1-year-old face (derived with bootstrapping¹⁷) was automatically placed onto each target image by a 3-D surface registration algorithm. This process gradually warps the shape of the template into the shape of the target face,^18,19 thus sampling each face at corresponding locations across the entire surface. Each point configuration was made symmetrical.¹⁷ The location, orientation, and size of all point configurations were standardized using generalized Procrustes analysis.²⁰ Fitting of the template was performed in Mevislab, a platform for medical image visualization and developing image processing algorithms (http://www.mevislab.de), using custom-built modules developed at Katholieke Universiteit Leuven, Leuven, Belgium.¹⁹

Regression covariates included risk factors known to be, or that could plausibly be, associated with craniofacial shape: child’s sex, which is known to be associated with early craniofacial dimorphism¹⁶; maternal age, a risk factor for malformations²¹; and maternal smoking in pregnancy, a risk factor for malformations of the head and face.²² The child’s birth weight and maternal prepregnancy body mass index were included as factors potentially resulting in greater fat deposition around the cheeks.

To gauge individual variation in alcohol metabolism, mothers were asked how quickly they felt the effects of alcohol (very slowly or slowly, normally, or very quickly or quickly). This variable was considered as a potential modifier of any association between PAE and craniofacial shape. Although the measure is subjective, self-perceived alcohol effects may serve as a proxy variable, coding for unknown, complex, polygenic determinants of alcohol metabolism.

The distribution of PAE groups and covariates was described using frequency counts and percentages if categorical, and mean (SD) values if continuous. To test for a difference in craniofacial shape between the control and each PAE group in each tier, partial least squares regression models were fitted. Partial least squares regression was chosen because the point configurations are highly collinear and because there are more point coordinates than observations (faces). All models included PAE group (coded as control = 0 and PAE = 1) to model the association with PAE. The predictors listed were controlled for by including them in each partial least squares regression model. A second analysis computed the same regression models to examine any association between craniofacial differences and PAE, stratified according to self-perceived alcohol effects.

At the global and regional level, the statistical association that controlled for covariates (partial R²) was computed. Statistical significance was computed at the global level using a permutation test with 1000 permutations.^23(p278)P < .05 was considered significant. Given that analysis of almost 70 000 points results in mean differences that vary across the face, craniofacial anatomical differences associated with PAE at each point were plotted using false colormaps, which show a generic face with each point indexed in color to the amount of difference and highlighting which regions were more changed or less changed.^11,24 Partial R², total anatomical difference, and difference in lateral and vertical directions and depth were plotted using separate colormaps.

For partial least squares regression and permutation testing, we used a custom-written code (by H.M.) in the Python programming language. The figures were generated using the Mayavi 3D visualization library (http://code.enthought.com/projects/mayavi/).

The characteristics of the children included in the analysis are summarized in Table 1. The cohort included 220 boys and 195 girls with a mean (SD) age of 363.0 (8.3) days at imaging. Of the 326 children with PAE, 133 (40.8%) were exposed in the first trimester only and 193 (59.2%) throughout the pregnancy.

There was no significant association between PAE and craniofacial shape at the global level (Table 2). There were, however, regional mean differences in craniofacial shape of children exposed to any alcohol (tier 1), regardless of whether PAE occurred in the first trimester only or throughout pregnancy (tier 2) (Figure 1). Regions of difference were concentrated around the midface, nose, lips, and eyes. Directional visualization showed that these differences corresponded to a general recession of the midface (Figure 1; blue areas in the depth section) and a superior displacement of points of the nose, especially the tip of the nose (Figure 1; red areas in the vertical section), indicating a shortening of the nose and upturning of the nose tip. This craniofacial phenotype was evident in both tiers 1 and 2.

In tier 3, changes were most marked with moderate to high PAE in the first trimester around eyes, midface, chin, and parietal region and with binge-level PAE in the first trimester around the lower lip. More important, a consistent phenotype comprising a recessed midface and upturned nose was evident in all exposure categories except in those with binge-level exposure in the first trimester only, where the recession was confined to the chin (retrognathia). Overall, point differences reached a maximum of 1.2 mm between groups.

The partial least squares regression analysis was repeated, stratified by maternal report of feeling the effects of alcohol normally and very quickly or quickly (omitting those who reported feeling the effects of alcohol slowly or very slowly owing to small group numbers). For children of those who felt the effects of alcohol normally, there was no significant global association in any exposure category (Table 3). Overall, the craniofacial phenotype was similar to that in the unstratified analysis but was more pronounced in the stratum of mothers who reported feeling the effect of alcohol very quickly or quickly. There was also a significant association at the global level in those who drank throughout in the tier 2 and tier 3 analyses (with the exception of the binge-level category). Regional associations with PAE (Figure 2) reached a difference of 1.6 mm in the forehead and midface in the stratum of mothers who reported feeling the effect of alcohol very quickly or quickly.

There were significant global differences in craniofacial shape associated with maternal age, prepregnancy body mass index, children’s birth weight, and child’s sex. Maternal smoking during pregnancy was not associated with craniofacial shape at a global level. The tier 1 regression model showing these associations is in eTable in the Supplement (global associations) and eFigure in the Supplement (regional associations).

To our knowledge, this study is the first to examine the association between the face of the child and common patterns of PAE, using objective, holistic methods of craniofacial phenotyping. A consistent association with craniofacial shape was observed in almost all exposure groups, with differences concentrated on regions around the nose, eyes, upper lips, and lower lips. Results indicate a mild midfacial recession suggestive of subclinical hypoplasia and an upturning of the nasal tip in those exposed to alcohol prenatally. This phenotype was evident even when drinking was at a low level and mothers ceased alcohol consumption in the first trimester.

In the unstratified analysis, we observed the strongest association with facial shape in children of mothers who drank at moderate levels in the first trimester only. In the stratified analysis, craniofacial differences were strongest in children of mothers who said they felt the effects of alcohol quickly, particularly if they continued drinking throughout pregnancy and initially drank at moderate levels. The apparent discrepancy between the 2 analyses may be because fewer women who reported feeling the effects of alcohol quickly drank throughout pregnancy.¹⁵ A higher proportion of mothers who felt the effects of alcohol normally in those exposure groups may have masked the association in the unstratified analysis.

The PAE group with binge-level alcohol consumption comprised 50% to 70% of mothers with infrequent or single-binge exposures, while in 20% to 30% of mothers, the binge exposure occurred weekly or more often. This heterogeneity may explain why we did not see a strong association between binge-level PAE and craniofacial shape.

Facial abnormalities have previously been described in children with PAE who do not meet diagnostic criteria for FAS. Suttie et al⁵ analyzed 69 children whose mothers consumed up to 13 standard drinks (14 g of AA) per week or drank at binge levels and found that 28 children showed patterns of facial anomalies similar to those seen in FAS or partial FAS. These anomalies were also associated with cognitive impairments. In their group comparisons, those authors noted an elevation of the lower portion of the nose in those with FAS, as well as smoothing of the philtrum, reduced palpebral fissure length, midfacial hypoplasia, and retrognathia. The only previous study to examine low levels of PAE found that 66% of 79 newborns in the exposure group had some facial abnormality.²⁵ However, observed abnormalities were not objectively defined and assessed by unblinded observers. Moreover, the weekly mean intake of alcohol per trimester was reported retrospectively, and it is not possible to know if there were spikes in consumption at particular time points, including the first trimester and the period before recognition of pregnancy. Our study has examined the association of low levels of PAE using a more rigorous assessment of both face shape and alcohol consumption. We show that certain aspects of the phenotype (upturning of the nasal tip and midfacial hypoplasia) can be detected even when the maximum number of standard drinks (10 g of AA) did not exceed 7 per week and 2 per occasion. We did not observe the classic diagnostic features for FAS of smooth philtrum, reduced palpebral fissure length, and thin vermillion of the upper lip; it is likely that these features emerge only at higher levels of exposure.

During embryogenesis, facial bone and cartilage are derived from the cranial neural crest. Sizing and positioning of facial elements begins 17 to 18 days after fertilization and before most pregnancies are recognized.²⁶ Evidence from mouse studies shows that exposure to ethanol affects all stages of neural crest development, resulting in variation in craniofacial appearance, depending on the gestational timing of exposure. For example, alcohol exposure at gestational day 7 (the 15th-17th day in human development²⁷) leads to severe midfacial hypoplasia, an elongated upper lip, and a deficient philtrum, while exposure at day 8.5 causes mild midfacial hypoplasia, a shortened upper lip, and a preserved philtrum.²⁸ We observed a similar facial phenotype to that seen in animal models, particularly after first-trimester moderate PAE. Although it was not possible to localize the timing of exposure as precisely as in these animal studies, our findings confirm an association between moderate PAE and facial shape in the first trimester in humans, which is convergent with the animal evidence.

Even in children of mothers who drink heavily, facial abnormalities associated with PAE are highly variable,⁵ which may reflect variation in the timing of exposure in the first trimester or unmeasured risk factors. For example, genetic variants in maternal and fetal alcohol metabolism have been shown to influence the level of alcohol and/or its toxic metabolites experienced by the fetus. Mendelian randomization studies using the Avon Longitudinal Study of Parents and Children predicted different patterns of PAE and variable outcomes in offspring depending on several variants in the alcohol dehydrogenase gene in the mother or child (eg, ADH1B [OMIM 103720], predictive of reduced alcohol consumption and associated with higher academic achievement in offspring).^29,30 Furthermore, regarding the direct role of genes involved in alcohol metabolism in modifying risk, evidence from laboratory studies using ethanol-sensitive and ethanol-resistant chickens, mice, and zebra fish provides insight into the multifactorial genetics of ethanol-mediated cell signaling disruption and neural crest apoptosis.³¹

We observed that children of mothers who reported feeling the effects of alcohol quickly or very quickly exhibited larger craniofacial differences in most exposure groups. We hypothesize that this rate of feeling the effects of alcohol reflects genetically determined variation in alcohol metabolism. The rate of feeling the effects of alcohol may ultimately be clinically useful to differentiate individuals with a greater susceptibility to the effects of alcohol, conferring increased fetal vulnerability, and may in part explain the heterogeneity of outcomes in alcohol-exposed pregnancies and in FASD.

Craniofacial development closely corresponds to brain induction and expansion, and, as such, characteristic facial differences have been linked to brain abnormalities and cognitive outcome in FASD.³¹ Correlative face-brain phenotypes have been described in human and animal studies, suggesting that the type and severity of brain abnormality may be predicted in part by hypoplasia of the midface,^28,32 and that the classic facial features of FASD (short palpebral fissure, smooth philtrum, and thin upper lip) may assist in identifying children at risk of developing neurobehavioral deficits.⁵ Given that cognitive outcomes for the children in our study have not yet been examined in this context, it is as yet unknown if the craniofacial differences found are of diagnostic or predictive value.

This study is a well-described cohort of mother-child dyads with detailed PAE data and classification not available in many other studies. The method of craniofacial phenotypic analysis used in this study is the most objective and sensitive available, to our knowledge. Although the magnitude of the association was variable, we observed very similar anatomical differences using almost every PAE categorization, which essentially constitutes several replications of the findings and demonstrates their robustness. A link between these facial changes and brain structure and functioning remains to be investigated; in the meantime, we provide additional evidence for an association between alcohol consumption and the developing fetus.

We did not observe the auricular anomalies previously documented with low-level PAE in other studies.⁴ Owing to the geometry of the multiple cameras used for imaging and the anatomy of the ear, it was not possible to capture high-quality images of the ear, and the analysis may have been obscured by photographic noise in this region.

We postulate that the rate of feeling the effects of alcohol is a proxy for metabolic factors influencing PAE and its association with facial shape, but have no direct measure such as blood alcohol concentration or alcohol elimination rates to examine. Investigation of maternal and/or child allelic differences (genotypes) at specific genes associated with alcohol metabolism or alcohol use behavior is under way.

The results of this study suggest that even low levels of alcohol consumption can influence craniofacial development of the fetus and confirm that the first trimester is a critical period. We observed aspects of a craniofacial phenotype with almost any level of PAE, something previously only documented following a high level of long-term alcohol exposure. Although the clinical significance of our findings is yet to be determined, these findings support the conclusion that, for women who are, or may become pregnant, avoiding alcohol is the safest option.

Accepted for Publication: March 2, 2017.

Corresponding Author: Jane Halliday, PhD, Public Health Genetics, Murdoch Childrens Research Institute, 50 Flemington Rd, Parkville, Victoria 3052, Australia (jane.halliday.h@mcri.edu.au).

Published Online: June 5, 2017. doi:10.1001/jamapediatrics.2017.0778

Author Contributions: Ms Muggli and Mr Matthews are co-first authors. Ms Muggli and Mr Matthews had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Muggli, Penington, O’Leary, Forster, Donath, Anderson, Nagle, Craig, Elliott, Halliday.

Acquisition, analysis, or interpretation of data: Muggli, Matthews, Claes, O’Leary, Forster, Donath, Lewis, Nagle, White, Elliott, Halliday.

Drafting of the manuscript: Muggli, Matthews, Donath, Nagle, White, Elliott, Halliday.

Critical revision of the manuscript for important intellectual content: Muggli, Matthews, Penington, Claes, O’Leary, Forster, Donath, Anderson, Lewis, Nagle, Craig, Elliott, Halliday.

Statistical analysis: Matthews, Claes, Donath, Halliday.

Obtained funding: Muggli, O’Leary, Forster, Anderson, Nagle, Craig, Elliott, Halliday.

Administrative, technical, or material support: Muggli, Matthews, Nagle, Craig, Elliott, Halliday.

Study supervision: Muggli, Penington, Claes, Anderson, Craig, Elliott, Halliday.

Conflict of Interest Disclosures: None reported.

Funding/Support: This work is supported by grant 1011070 from the Australian National Health and Medical Research Council, Senior Research Fellowships 1081288 (Dr Anderson) and 1021252 (Dr Halliday) from the Australian National Health and Medical Research Council, Practitioner Fellowship 1021480 (Dr Elliott) from the Australian National Health and Medical Research Council, and the Victorian State Government’s Operational Infrastructure Support Program.

Role of Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Contributions: Ine Saey, MS, Department of Electrical Engineering, KU Leuven, contributed to the preliminary analyses; she was not compensated for her contribution. We also thank all the women and their children who are taking part in this study and the medical photographers who undertook the 3-dimensional imaging.