In this study, Avy
was used primarily as a sensitive reporter of epigenetic changes in response to maternal ethanol consumption. The C57BL/6J mouse is null (a
) at the Agouti
locus, so it has a black coat color. Avy
is a gain-of-function, semi-dominant mutation and so the coat color of heterozygous (Avy/a
) mice in the C57BL/6J background is a direct read out of Avy
transcriptional activity and DNA methylation. The nature of the matings used in this study, an Avy/a
male crossed with an a/a
female, means that only 50% of the offspring will inherit the Avy
allele and be useful for coat color phenotyping. The remaining (a/a
) offspring will be black. To study the effects of gestational ethanol exposure, female a/a
C57BL/6J mice were supplied with 10% (v/v) ethanol in their drink bottles for eight days after fertilization by a congenic male carrying the Avy
46 litters, 242 total offspring, 109 Avy/a
offspring). To evaluate the effects of preconceptional ethanol exposure, female a/a
mice were given 10% (v/v) ethanol for four days per week for ten weeks prior to fertilization (n
22 litters, 131 total offspring, 69 Avy/a
offspring). The Avy
allele was passed through the male germ line to avoid the bias associated with maternal transmission, where epigenetic marks can be incompletely cleared between generations 
. Control mice were given water instead of ethanol (n
37 litters, 189 total offspring, 91 Avy/a
offspring). Maternal ethanol exposure during gestation did not significantly alter Mendelian inheritance of the Avy
allele (data not shown) or litter size (control 5.1±0.4, ethanol exposed 5.2±0.3, mean±SEM, Student's t
The establishment of epigenetic marks at Avy
occurs during early embryogenesis and is a probabilistic event. The resulting variable expression of Avy
among genetically identical mice produces individuals with a predictable range of coat colors. We found that, in the absence of any treatment, 21% of the offspring of Avy/a
sires were yellow, 66% were mottled and 13% were pseudoagouti (). Gestational ethanol exposure resulted in a higher proportion of pseudoagouti (Pearson's chi-square test, p<0.05). Twenty-eight percent of offspring were pseudoagouti compared with 13% in the control group (). Preconceptional ethanol exposure produced a similar trend (Pearson's chi-square test, p<0.05). This shows that ethanol exposure can influence the establishment of Avy
expression early in development. It increases the probability of transcriptional silencing at this particular locus. To confirm that the coat color correlated with DNA methylation at the Avy
allele in gestationally exposed mice, 11 CpG dinucleotides in the LTR cryptic promoter of the Avy
IAP were subjected to bisulfite sequencing (). The results showed that, as expected, ethanol-exposed yellow mice were hypomethylated compared to ethanol-exposed pseudoagouti mice. Interestingly, atypical hypermethylated clones were found in five out of six yellow mice in the ethanol-exposed group, but they were clearly not sufficient to affect coat color. In the ethanol-exposed group 11% of the CpG dinucleotides were methylated compared to 2% in the control group. Using this measure a Student's t
-test or non-parametric equivalent was unsuitable because the data did not meet the distribution requirements of being spread on a continuum. So we analyzed allele-specific methylation. In the ethanol-exposed group 23% of clones showed evidence of methylation, n
91, compared with 8% of clones in the control group, n
71 (Pearson's chi-square tests, p<0.01). In contrast, total DNA methylation level in ethanol exposed pseudoagouti mice (61%) was not significantly different to that observed in the controls (65%, Student's t
0.27). Equivalent results were obtained from a random effects model which allowed for the clustering of clones within mice (p
0.23). Bisulfite sequencing was also carried out on control and ethanol-exposed mottled mice and we found the results extremely variable from one mouse to the next within both groups (Figure S1
). Presumably, this is the result of the small size of the tissue sample (tail tip). The variegated expression in mottled mice means that any one sample could represent only one clonal patch, which could harbour an active or an inactive Avy
allele and not represent the true methylation state of the whole animal. For this reason mottled mice were not used in our analyses.
Gestational and preconceptional ethanol exposure produced a higher proportion of pseudoagouti Avy mice.
Avy methylation in control offspring and offspring exposed to ethanol in utero.
The effects of Avy
expression are pleiotropic. For example, yellow mice exhibit hyperphagia, hyperglycemia, non-insulin-dependent diabetes and adult onset obesity 
. We did not assay these other phenotypes following ethanol exposure in our mice as their relevance to humans is questionable since no human ortholog of the Avy
allele has been identified. So, while Avy
was initially useful as a sensitive indicator of epigenetic changes, any further study of FAS-like phenotypes must necessarily focus elsewhere in the genome. For this reason and the fact that variable Agouti
expression would confound many phenotypes, all subsequent analyses were performed on the congenic a/a
siblings of the Avy
mice. In the mouse, IAPs are present at approximately 1,000 copies per haploid genome 
. To see if gestational ethanol exposure changed the methylation level of IAPs globally, we performed bisulfite sequencing using PCR primers that anneal to all IAP LTRs and analyzed ten CpG sites in both the tail and forebrains of a/a
mice. Tail DNA from eight ethanol-exposed mice (66 clones total) and eight control mice (65 clones total) were compared, and forebrain DNA from five ethanol-exposed mice (33 clones total) and five control mice (43 clones total) were compared using the Student's t
-test. All samples were highly methylated and no differences between the ethanol exposure group and controls were detected (Figure S2
). This suggests that only a subset of IAPs, perhaps those that are usually hypomethylated, are sensitive to ethanol exposure.
To detect changes in gene expression genome-wide, we performed expression arrays with liver tissue. The benefit of using liver is its homogeneity; it consists mainly of hepatocytes and consequently subtle changes will be detectable. We compared gene expression between age-matched male mice from the gestational ethanol group (three samples) and controls (four samples). In addition to inter-individual variation, some of the genes were consistently differentially expressed in the ethanol-exposed group (Table S1
). Twelve genes were significantly down-regulated (p<0.05) in the ethanol-exposed mice. Three of these; LIM domain and actin binding 1
), also known as Eplin
, Suppressor of cytokine signaling 2
) and CDK5 and Abl enzyme substrate 1(Cables1)
have been associated with growth 
. Three; Socs2
, Very low density lipoprotein receptor
) and Cables1
have been associated with development of the nervous system 
and one, Hepcidin antimicrobial peptide
), has been reported to be down-regulated in the livers of alcohol-fed rats 
We next focused on identifying the characteristic features of FAS in a/a
pups exposed to moderate levels of alcohol in utero
. All pups (from first litters) were weighed at three weeks of age. It was particularly important not to study Avy
mice in these experiments because of the effects on body weight due to Agouti
expression. Because litter size is known to influence body weight at weaning, we initially restricted our analysis to litters of 4–5 pups. The gestational ethanol exposure group consisted of 22 offspring, while the control group consisted of 26 offspring. The results () show that the mean weight of offspring of dams that consumed ethanol were significantly lower than that of controls (Student's t
-test, p<0.05). A second analysis included litter size as a random effect. Analysis of Variance of weight at 3 weeks, after adjustment for litter size, confirmed that the mean weight of the ethanol exposed group (n
73) was statistically significantly smaller than the mean weight of the controls (n
44, p<0.001, data not shown).
Offspring in the gestational ethanol exposure group have a statistically significantly lower mean weight than the control group (Student's t-test, p<0.05).
The heads of 28–30 day old a/a mice (seven mice from gestational ethanol exposure group and 10 control mice) were subjected to micro-computed tomography, and three-dimensional computer-reconstructions at 18 µm resolution were made of each skull. Visual inspection of the reconstructions revealed an obviously smaller skull size in the ethanol group compared to controls. In addition, differences in shape in a few, but not all, individuals in the ethanol group were apparent. Most notable was the marked leftward deviation of the midface in one male () and a significantly reduced interfrontal bone in one female (). To provide more quantitative information on skull shape, the 3D co-ordinates of thirty-four landmarks were recorded for each skull and used in various mathematical-based shape and form analyses.
Variable midfacial dysmorphism and microcephaly in a/a offspring of mothers that consumed ethanol during gestation.
There are two classic approaches in geometric morphometrics: superimposition based methods such as Generalized Procrustes Analysis (GPA) 
or invariant analyses of shape, such as Euclidean Distance Matrix Analysis (EDMA) 
. GPA involves translation, rotation and scaling of landmark data through an iterative process during which the distances between the shapes are minimized by applying least-squares criteria. We used GPA to test for the mean shape difference between the groups and to quantify and visualize localized differences in the cranial shape. We also applied the multivariate ordination method Canonical Variates Analyses to the output of the GPA. EDMA, in contrast, uses a coordinate-free (or invariant) approach in which all the landmarks are converted into a matrix of inter-landmark distances 
. We used EDMA to find the landmark pairs that show the most difference between two groups.
Analysis of skull centroid sizes confirmed the observations of statistically significantly reduced cranial size in the ethanol group, even when the smaller body weight is taken into account (ANOVA, p<0.05; ). Although the severe leftward deviation of the one male skull is biologically highly relevant, we chose to exclude this sample from subsequent shape analyses because of its significant impact on the results. This permitted us to assess the significance of other more subtle changes. However, it was included in the univariate analyses of relative cranial dimensions. In the absence of this outlier, CVA still revealed greater variation in overall craniofacial shape within the ethanol group (). Canonical variate 1 (CV1) clearly separated the females in the ethanol group from other skulls, suggesting a more pronounced effect on female skull shape. Notably, all the females as well as the included males from the ethanol group had positive values for CV2, whereas the controls spanned both negative and positive values (see ), indicative of a similar trend in shape alteration in response to this level of gestational exposure to ethanol. One female from the ethanol group appeared to be unaffected in terms of craniofacial shape and grouped with the control females in all analyses.
Quantitative analysis of the effects of gestational exposure to ethanol on skull shape.
We then used EDMA to assess the differences in form between the ethanol and control groups. Analysis of the 561 possible inter-landmark measurement combinations assessed by the 34 assigned landmarks (i.e. 34(34-1)/2) demonstrate that the majority show a consistent ratio below one, indicative of the fact that they are changed only relative to skull size and do not reflect localized altered shape. Nevertheless, numerous inter-landmark measures were shown to be significantly different from this mean form. The twenty most significant differences (α
0.1) in either direction from the mean form are shown in . Strikingly, almost all of these forty most significant differences pertain to midfacial and palatal inter-landmark measures, highlighting the sensitivity of this region to the ethanol. In particular, these data reveal that the ethanol group as a whole have a relatively wider inter-orbital distance (inter-landmark measure 7–11), yet relatively shorter midface than controls (reflected in multiple inter-landmark measures). This is consistent with the CVA findings.
A univariate analysis of inter-landmark distances (normalized to centroid size) also supported these differences between the ethanol and control groups, in particular, confirming the greater relative cranial and inter-orbital width in both males and females compared to the sex-matched controls (). Females from the ethanol group also showed greater variation in ‘nare’ height (data not shown), while males from the ethanol group showed reduced rostrum length (). Although less severe than the changes found with acute ethanol exposure in the mouse 
, many of these differences are reminiscent of the facial changes seen in individuals presenting the milder end of fetal alcohol spectrum disorder, and support the notion that this ethanol regime provides a useful and relevant model for the effects of ethanol intake in humans.