Reprogramming of genetic networks during initiation of the Fetal Alcohol Syndrome


  • Maia L. Green,

    1. Department of Pathology, Anatomy and Cell Biology, Thomas Jefferson University, Philadelphia, Pennsylvania
    2. Department of Molecular, Cellular and Craniofacial Biology, University of Louisville, Louisville, Kentucky
    Search for more papers by this author
  • Amar V. Singh,

    1. Department of Pathology, Anatomy and Cell Biology, Thomas Jefferson University, Philadelphia, Pennsylvania
    2. Department of Molecular, Cellular and Craniofacial Biology, University of Louisville, Louisville, Kentucky
    Search for more papers by this author
  • Yihzi Zhang,

    1. Department of Molecular, Cellular and Craniofacial Biology, University of Louisville, Louisville, Kentucky
    Search for more papers by this author
  • Kimberly A. Nemeth,

    1. Department of Pathology, Anatomy and Cell Biology, Thomas Jefferson University, Philadelphia, Pennsylvania
    Current affiliation:
    1. FemCare Product Safety and Regulatory Affairs, The Procter & Gamble Company, Cincinnati, OH 45224
    Search for more papers by this author
  • Kathleen K. Sulik,

    1. Department of Cell and Developmental Biology, University of North Carolina, Chapel Hill, North Carolina
    Search for more papers by this author
  • Thomas B. Knudsen

    Corresponding author
    1. Department of Pathology, Anatomy and Cell Biology, Thomas Jefferson University, Philadelphia, Pennsylvania
    2. Department of Molecular, Cellular and Craniofacial Biology, University of Louisville, Louisville, Kentucky
    • University of Louisville, School of Dentistry, 501 South Preston Street Louisville, KY, 40202
    Search for more papers by this author

  • GEO dataset (GDS): GSE1074.


Fetal Alcohol Spectrum Disorders (FASD) are birth defects that result from maternal alcohol use. We used a non a priori approach to prioritize candidate pathways during alcohol-induced teratogenicity in early mouse embryos. Two C57BL/6 substrains (B6J, B6N) served as the basis for study. Dosing pregnant dams with alcohol (2× 2.9 g/kg ethanol spaced 4 hr on day 8) induced FASD in B6J at a higher incidence than B6N embryos. Counter-exposure to PK11195 (4 mg/kg) significantly protected B6J embryos but slightly promoted FASD in B6N embryos. Microarray transcript profiling was performed on the embryonic headfold 3 hr after the first maternal alcohol injection (GEO data series accession GSE1074). This analysis revealed metabolic and cellular reprogramming that was substrain-specific and/or PK11195-dependent. Mapping ethanol-responsive KEGG (Kyoto Encyclopedia of Genes and Genomes) pathways revealed down-regulation of ribosomal proteins and proteasome, and up-regulation of glycolysis and pentose phosphate pathway in B6N embryos; and significant up-regulation of tight junction, focal adhesion, adherens junction, and regulation of the actin cytoskeleton (and near-significant up-regulation of Wnt signaling and apoptosis) pathways in both substrains. Expression networks constructed computationally from these altered genes identified entry points for EtOH at several hubs (MAPK1, ALDH3A2, CD14, PFKM, TNFRSF1A, RPS6, IGF1, EGFR, PTEN) and for PK11195 at AKT1. Our findings are consistent with the growing view that developmental exposure to alcohol alters common signaling pathways linking receptor activation to cytoskeletal reorganization. The programmatic shift in cell motility and metabolic capacity further implies cell signals and responses that are integrated by the mitochondrial recognition site for PK11195. Developmental Dynamics 236:613–631, 2007. © 2007 Wiley-Liss, Inc.


Prenatal alcohol damage affects an estimated 1% of liveborn infants (May and Gossage,2001) and is the leading preventable cause of mental retardation (Sokol et al.,2003). Although publicity campaigns have stirred women to stop alcohol consumption once they recognize that they are pregnant, national survey data indicate that 13% of women continue to use alcohol during pregnancy (see Bertrand et al.,2004). Fetal Alcohol Spectrum Disorders (FASD) refers to the full spectrum of birth defects caused by prenatal alcohol exposure. The Fetal Alcohol Syndrome (FAS) comprises a subset of FASD, including a consistent pattern of physical abnormalities most evident in the face, with intrauterine growth retardation and neuro-developmental problems (Jones et al.,1973; Webster et al.,1983; Bertrand et al.,2004). Developmental stage at the time of prenatal alcohol exposure, coupled with differing patterns and amounts of alcohol consumption by the mother, account for some of the known variability and uncertainty in ethanol-induced teratogenic end points. Animal studies as well as a limited amount of human data also show that maternal genotype is a key player (Viljoen et al.,2001; Goodlett et al.,2005). For these reasons and more, the U.S. Surgeon General recommends that pregnant women and women who may become pregnant abstain from alcohol consumption (

The complex etiology of FASD requires basic research to improve mechanistic understanding of prenatal alcohol damage and clinical research to identify at-risk newborns at the earliest possible age (Streissguth et al.,1996; Bearer et al.,2005; Green et al.,2005; Sulik,2005). Based on mechanistic intervention models, candidates for critical effects of prenatal alcohol exposure include: secondary zinc deficiency preventable with vitamin supplementation (Carey et al.,2003); induction of excessive apoptosis or possible excitotoxic bioactivity in the developing neural system that can be protected with neurotrophic peptides (Wilkemeyer et al.,2003; Toso et al.,2006); interference with L1 adhesion molecule (L1 CAM) function that can be prevented with long-chain alcohols such as n-octanol (Chen et al.,2005); and oxidative stress and ROS generation that can be prevented with a superoxide dismutase/catalase mimetic (EUK-134) (Chen et al.,2004) or nicotinamide (Ieraci and Herrera2006). Alcohol-related changes in gene expression have been reported for several developmentally important loci (Lee et al.,1997; Rifas et al.,1997; Ahlgren et al.,2002; Poggi et al.,2003; da Lee et al.,2004; Xu et al.,2005; Yamada et al.,2005) and functional gene categories (Rahman and Miles,2001; Saito et al.,2002; Gutala et al.,2004; da Lee et al.,2004; Kerns et al.,2005; Miller et al.,2006). Regardless of this detailed molecular information, the complex and often diverse responses of cells and tissues to alcohol under different experimental conditions have complicated efforts to interpret a response signature for alcohol-related birth defects. Furthermore, alcohol may alter gene expression at multiple levels over time, including transcription, mRNA stability, protein synthesis, and posttranslational modification (Lang et al.,2000; Martin et al.,2004; Miller et al.,2006). This motivates an integrative systems-based approach to link genomic response with genetic susceptibility of the sensitive target tissue.

The present study used a non a priori approach to assess molecular phenotype in the cranial neural folds (headfold) of early mouse embryos soon after maternal alcohol treatment. Two related lineages of C57BL/6 mice with apparent differential susceptibility to alcohol served as the basis for study. On the one hand, C57BL/6J (B6J) pregnancies exposed by two injections of 2.9 g/kg ethanol spaced 4 hr apart represent an established model of FAS, simulating a heavy binge-like episode (Webster et al.,1983; Kotch and Sulik,1992). Here we show that the closely related C57BL/6NCrl substrain (B6N) is far less susceptible to FAS and, furthermore, observe that the differential risk for FAS can be modulated pharmacologically with PK11195, a potent ligand with specific binding to the mitochondrial peripheral-type benzodiazepine receptor recognition site. PK11195 has been shown to protect early mouse embryos from eye and brain defects induced with diverse teratogens (O'Hara et al.,2002, 2003; Charlap et al.,2003), and to protect adult tissues from some inflammatory lesions (Waterfield et al.,1999; Bribes et al.,2002, 2003a, b). In contrast to PK11195-mediated protection of alcohol-induced malformations in B6J embryos, the effect of PK11195 in B6N embryos increased the risk of FAS. We exploited these differences to characterize the genomic responses to alcohol on day 8 of gestation. Microarray transcript profiling of the embryonic headfold at 3.0 hr after alcohol exposure or PK11195 counter-exposure enabled prioritization of candidate pathways that integrate the genomic response with genetic susceptibility of the system.


Substrain Differences in Susceptibility to Alcohol-Related Birth Defects

Several (3–4) litters of each strain were used to verify craniofacial dysmorphology induced with acute alcohol exposure on day 8 of gestation. Previous studies have shown that in comparison to the B6J strain, outbred CD-1 mice take higher ethanol concentrations and longer exposure times for comparable incidence rates of defects (Hunter et al.,1994; Kotch et al.,1995; Chen et al.,2000). The range of sensitivity in B6N mice was unknown to us at the onset of this study. A preliminary analysis was run with two injections of 2.9 g/kg alcohol spaced 3 or 4 hr apart. At day 11, the B6J litters had many severely retarded embryos and 20–25% of viable embryos displayed craniofacial defects such as microphthalmia, exencephaly (Fig. 1), small cerebral hemispheres, and facial dysmorphology. Other adverse outcomes included pericardial edema and lumbar spina bifida oculta. These malformations were consistent with expectations from previous studies (Sulik et al.,1986; Kotch and Sulik,1992). Histological analysis at day 14 confirmed disruption or delays in development of the eye, forebrain, and orofacial region (not shown). Malformations were not observed in the exposed CD-1 or B6N litters (Fig. 1).

Figure 1.

Alcohol-induced dysmorphogenesis of the developing face and brain of mouse embryos. Scanning electron micrographs of day-11 embryos comparing control (vehicle) CD-1 (A) and B6N (B) embryos with ethanol-exposed embryos from B6N (C) and B6J (D) stocks. Examination of several litters of the B6N stock revealed no demonstrable dysmorphogenesis. The B6J stock showed defects consistent with alcohol-related birth defects following exposure to two injections of 2.9 g/kg ethanol, 4.0 hr apart on day 8 of gestation. An anterior neural tube closure defect is evident in this particular embryo.

A teratological evaluation was conducted with B6J and B6N pregnancies (Table 1). One subset of dams was exposed to alcohol (or control) and a second subset received alcohol in conjunction with 4 mg/kg PK11195 (or PK11195 alone). Maternal weight averaged 23.8 ± 1.4 g (n = 49) in B6N dams versus 21.6 ± 1.5 g (n = 41) in B6J dams at the time of breeding (day 0). Otherwise, non-alcohol-treated litters (control, PK11195) were similar between B6J and B6N substrains for litter size (8.1 implants per litter), fetal weight (0.75 g per fetus), resorption rate (10–13% embryo losses), and spontaneous malformation incidence (5% occurrence of small eye).

Table 1. Differential Responses of B6N and B6J Litters to Ethanol Exposure on Day 8a
  • a

    Significant by two-way ANOVA for background (◂), treatment (▸), or interaction (◃▹) at *P < 0.05, **P < 0.01, ***P < 0.001. When ANOVA identified a significant group difference, posthoc analysis was performed by Bonferroni-corrected t-test comparing the alcohol group with corresponding control (▴) or cotreated (▾) groups in each substrain, or comparing corresponding control (▵) or cotreated (▿) groups across substrains. CRL, crown-rump length; CRC, circumference of neurocranium; ERDs, eye reduction defects (see Methods section for meaning of score).

  • Mean effects per litter ± SE on day 17 of gestation.

Number of litters (n)141212111510610
Fetuses/implants (#)92/10287/9889/10182/98103/11749/8540/4667/85
Resorption incidence (%)9.9 ± 3.513.7 ± 8.614.8 ± 7.615.9 ± 6.313.0 ± 5.034.2 ± 9.212.3 ± 4.221.1 ± 10.4
Fetal weight (g) ▸***0.741 ± 0.0270.627 ± 0.037 ▴*0.758 ± 0.0200.679 ± 0.0340.768 ± 0.0220.646 ± 0.029 ▴*0.718 ± 0.0170.704 ± 0.015
CRL (mm) ▸**18.27 ± 0.2517.10 ± 0.40 ▴*18.47 ± 0.2917.58 ± 0.3418.57 ± 0.2717.60 ± 0.4017.75 ± 0.3017.81 ± 0.18
CRC (mm) ◂***, ▸***19.47 ± 0.2518.43 ± 0.34 ▴*19.61 ± 0.1718.79 ± 0.3320.54 ± 0.1619.10 ± 0.46 ▴*19.65 ± 0.1319.70 ± 0.11
CRC/CRL ◂***1.067 ± 0.007 ▵*1.080 ± 0.0121.064 ± 0.014 ▿*1.070 ± 0.0111.107 ± 0.009 ▵*1.086 ± 0.0141.120 ± 0.019 ▿*1.106 ± 0.010
Grossly malformed (%) ▸***, ◃▹***7.1 ± 3.810.7 ± 4.13.1 ± 1.617.8 ± 6.21.7 ± 1.139.6 ± 9.1 ▴***, ▾**9.5 ± 7.113.1 ± 3.1
Eye malformations (%) ▸***, ◃▹*7.1 ± 3.89.8 ± 4.23.1 ± 1.616.8 ± 5.81.7 ± 1.132.0 ± 8.6 ▴***, ▾*13.1 ± 3.19.5 ± 7.1
ERDs (score) ▸**, ◃▹*0.116 ± 0.0620.158 ± 0.0820.046 ± 0.0280.227 ± 0.0800.017 ± 0.0110.751 ± 0.330 ▴***, ▾**0.107 ± 0.0820.131 ± 0.031

Resorption rates trended higher in alcohol-treated B6J litters (34% mean resorptions) than B6N litters (14% mean resorptions). PK11195 had a tendency to lessen the impact of alcohol in B6J litters (21% mean resorptions). Due to highly variable litter effects, the differences in resorption rates were not quite significant by ANOVA. Mean fetal weight, crown rump length (CRL), and neurocranial circumference (CRC) were all affected by alcohol in both substrains. Two-way ANOVA (treatment, substrain, interaction) identified significant treatment-related effects on these parameters that were attributed to alcohol exposure in either substrain. In addition, normal B6J fetuses showed a slightly larger (∼5% increase) mean CRC than B6N fetuses. This resulted in significant group differences across treatments and substrains. When we corrected CRC by CRL, only the substrain effect remained. This indicated a disproportionate effect of alcohol on fetal head growth that may reflect a natural tendency toward larger CRC in the B6J substrain at the time of evaluation on day 17.

Craniofacial malformations on day 17 were consistent with the preliminary analysis on days 11 and 14 and included mostly anterior NTDs (exencephaly, anencephaly), microcephaly, and midfacial hypoplasia in the B6J substrain (Fig. 2). The eye was the most evident target and the spectrum of eye reduction defects (ERDs) ranged from subtle (coloboma, narrowing of the pupillary ring) to degrees of microphthalmia and apparent complete anophthalmia (Fig. 3). Effects on gross malformations and ERDs were significant by two-way ANOVA for treatment (P < 0.001) and treatment × substrain interaction (P < 0.05), but not for substrain alone (Table 1). Post-analysis localized the significant effects to the B6J alcohol condition. Malformation rates were 39.6% (total malformed) and 32.0% (ERDs) in the B6J-alcohol condition versus 10.7 and 9.8%, respectively, in the B6N-alcohol condition. Therefore, the risk for alcohol-related birth defects was higher in B6J than B6N embryos.

Figure 2.

Alcohol-induced craniofacial symptoms. Formalin-fixed fetuses at day 17 following day-8 exposure, selected from various litters associated with data in Table 1. A: Normal fetus from B6N control group. B: Microcephalic fetus with midfacial hypoplasia from alcohol-treated B6J group. C,D: Exencephalic (C) and anencephalic (D) fetuses from B6N alcohol+PK11195 cotreatment group. [Color figure can be viewed in the online issue, which is available at]

Figure 3.

Alcohol-induced eye reduction defects (ERDs). Typical examples at day 17 following day-8 exposure, selected from various litters associated with data in Table 1. A: Normophthalmia (grade 0) fetus from B6N control group. B,C: Microphthalmia with coloboma, grade 2 (B) and severe microphthalmia with aphakia, grade 3 (C) from B6N alcohol+PK11195 cotreatment group. D: Apparent anophthalmia (grade 4) from alcohol-treated B6J group. [Color figure can be viewed in the online issue, which is available at]

B6J pregnancies did not show an increased risk for alcohol-related malformations in dams cotreated with PK11195 (Table 1). This protection was evident for ERDs by incidence (percentage) or graded effect (score) and was significant by Bonferroni-corrected t-test. And yet, despite the low rate of alcohol-related malformations among B6N litters, several dams that had been cotreated with alcohol and PK11195 had fetuses with quite severe FASD-like malformations (Fig. 2C,D). Therefore, although PK11195 cotreatment significantly lowered the risk for FASD in B6J pregnancies it tended to increase the risk in B6N pregnancies. Regarding dams exposed to PK11195 alone, spontaneous malformations trended lower in the B6N substrain versus the control group, but higher in the B6J substrain versus their respective control group (Table 1). Neither effect was statistically significant. The trend of malformations in the PK11195-alone group was affected by one B6J litter displaying fetal eye defects. Whether or not these trends in spontaneous malformation rates below statistical significance represent true biological effects cannot be ascertained from the design used here. In this experimental design (Table 1), we ran four groups (alcohol vs. control, B6J and B6N pregnancies) first followed by the remaining four conditions (EtOH-PK11195 vs. PK11195, B6J and B6N pregnancies). Analyzing the data by one-way ANOVA for the PK11195 and non-PK11195 groups independently gave essentially the same results as analyzing the whole set by two-way ANOVA; therefore, local variations in spontaneous background rates had no effect on the interpretation of the results. The differential response to PK11195 between substrains was not evident for resorptions, fetal weight reduction, CRC, or CRL. Because counter-exposure to PK11195 protected B6J fetuses on the one hand, but promoted these malformations in B6N fetuses on the other hand, we conclude that the risk for FASD is both substrain-dependent and PK11195-sensitive.

Gene Expression Profiling

To further characterize the differential response to alcohol and PK11195, we performed microarray-based analysis of transcripts in the headfold. Samples were collected 3.0 hr after maternal exposure to only the first injection of saline, alcohol, or alcohol+PK11195. Intactness of RNA was confirmed for all three treatment conditions (Fig. 4). We used two different platforms (PE [PerkinElmer] and AF [Affymetrix]) for this analysis, run as independent experiments. Pre-processed datasets after normalization had been subjected to statistical analysis to identify genes with significantly altered expression in either platform (see details in the Experimental Procedures section). This analysis returned 2340 (AF) and 628 (PE) probe elements with significantly altered expression (P ≤ 0.005). Probes passing the statistical (ANOVA) filter were set up as log2 expression ratios for exposed (alcohol vs. saline) and counter-exposed (cotreated vs. alcohol) conditions. Class comparison by unsupervised hierarchical clustering correctly paired the biological replicates within each platform (Fig. 5A,B); however, the overlap of annotated genes between platforms was only 55 genes. We therefore inquired as to whether the intrinsic patterns of the data were at all similar between the two datasets. Strength of the joint trends between platforms and replicates was assessed using the ordination technique of co-inertia analysis (CIA) (Culhane et al.,2003). CIA computes successive orthogonal axes with correspondence analysis and returns the percentage of total variance explained by each eigenvector to find the strongest trends in the co-structured datasets. The first pair of axes of ordination was selected, one from each platform, so as to be maximally co-variant and to represent the most important joint trend in the two platforms (Fig. 5C). Despite few overlapping genes, the internal data structure was concordant between AF and PE datasets as evidenced by similar inertia between platforms, conditions, and replicates. In this analysis, greater distance from the origin indicates the more discriminating trends for that treatment comparison. The least discriminating trend was displayed in the genetic response of alcohol-treated B6J embryos (sample A in Fig. 5C), which was the most susceptible in terms of FAS risk (Fig. 5D). The first (horizontal) axis of inertia resolved the B6N pattern (samples C vs. D in Fig. 5C), whereas, the second (vertical) axis resolved PK11195 counter-exposure (samples B vs. D in Fig. 5C). In comparing these trends in Figure 5C with FAS risk in Figure 5D, we find no simple association relating the genetic response of the headfold at 3.0 hr to genetic susceptibility of FAS risk.

Figure 4.

Quality control assessment with RNA 6000 Nano-LabChip on the Agilent 2100 Bioanalyzer. Independent samples of RNA isolated from the headfold of mouse embryos on day 8 of gestation. High quality mammalian total RNA based on the 2:1 ratio of 28S/18S rRNA bands was observed in all electropherograms. Lanes: L (ladder), headfold of saline treated (control) B6N embryos on day 8 of gestation (lane 1); headfold 3.0 hr after 2.9 g/kg ethanol exposure (lane 2); and headfold 3.0 hr after ethanol exposure and 4.0 mg/kg PK11195 cotreatment (lane 3). [Color figure can be viewed in the online issue, which is available at]

Figure 5.

Cross-platform comparison of AF and PE expression profiles. Cranial neural folds from day-8 mouse embryos were assayed 3.0 hr after maternal exposure to a single injection of saline (control), EtOH (alcohol), or EtOH-PK11195 (cotreated). Probes passing statistical (ANOVA) filter given as log2-transformed ratios for “E” exposed (alcohol-saline) and “CoT” counter-exposed (cotreated-alcohol) contrast. A,B: Class comparison of 2,340 probes and 628 probes screened by AF (A) and PE (B), respectively. Hierarchical clustering of replicate samples of B6J exposed (samples A1, A2) and counter-exposed (samples B1, B2) conditions, and of B6N exposed (samples C1, C2) and counter-exposed (samples D1, D2) conditions. The boxplot shows the log2 distribution of expression values for the probes in each of 8 conditions (line, median; box, 75%; whiskers, 95%; dots, outliers). The histogram gives the cumulative distribution of probes in each expression bin. C: The first two axes of ordination for AF and PE probes were projected for CIA. The horizontal axis reflects the first axis of the inertia and the vertical axis is the second axis of inertia. Circles (dots) and arrowheads (arrows) represent the projected coordinates of AF and PE datasets, respectively, with length of the line joining them being proportional to their divergence, e.g., shorter arrows indicate stronger joint trends between the platforms. Greater distance from the origin indicates the more discriminating trends for that group. Samples A–D refer to groups defined in Figure 5A and B. D: Corresponding contrast ratios for teratological data derived from Table 1 (log2 of mean ratio ± SE), for exposed (alcohol/control) of B6J (group A) and B6N (group C) substrains, and counter-exposed (cotreated/alcohol) of B6J (group B) and B6N (group D) conditions.

In the next step of the analysis, we evaluated the significant genes for their similarity to one another in expression patterns and functional annotation (Fig. 6). First, the filtered AF+PE genes were merged into a hybrid dataset of 2,906 non-redundant genes. Genes were then classified by expression profiles for the exposed and counter-exposed comparisons based on K-means clustering and Pearson correlation. Their distribution in each of 6 sets (k1 to k6) was as follows: 654 (k1), 342 (k2), 529 (k3), 500 (k4), 458 (k5), and 423 (k6). From these patterns, a set of simple rules was written to discriminate B6J and B6N responses to alcohol: membership to sets k2 and k5 implies B6J > B6N up-regulation; membership to set k3 implies B6J > B6N down-regulation; membership to sets k4 and k6 implies B6N > B6J up-regulation; and membership to set k1 implies B6N > B6J down-regulation. Annotation and functional classification used DAVID 2.1 beta ( The conversion of 2,906 gene records into UniGene identifiers returned 1,852 human and 2,317 murine records that mapped to 1,789 and 2,880, respectively, unique DAVID records. The highest ranking biological themes were stratified by enriched GO terms at level 4 (P ≤ 0.005). A gene-to-gene similarity matrix was generated to classify the highly-related genes into functionally related groups based on the three aspects of GO. Each expression cluster in Figure 6 contained 7–10 functional classifications and the annotated list is provided as an electronic supplement (see supplementary Table 1, which can be viewed at

Figure 6.

Functional mapping of 2,906 non-redundant genes altered in the embryonic headfold during maternal alcohol treatment. AF and PE probes passing the statistical filter (Fig. 5) were merged into a common dataset and the normalized data were partitioned into 6 clusters by K-means clustering using Pearson correlation. Numbers of genes in each cluster were: 654 (k1), 342 (k2), 529 (k3), 500 (k4), 458 (k5), and 423 (k6). Log2 expression ratios are displayed for replicate samples of B6J exposed (A1, A2) and counter-exposed (B1, B2) conditions, and of B6N exposed (C1, C2) and counter-exposed (D1, D2) conditions. Gene Ontology (GO) maps were developed from each cluster using Database for Annotation, Visualization, and Integrated Discovery ( to stratify functional groups by enriched GO terms (level 4, P ≤ 0.005). Abscissa of the heat map gives annotation terms and the ordinate gives functional groups. Functional groups are annotated in Supplementary Table 1.

We applied the EASEonline module ( to derive the significantly over-represented biological themes in the dataset. The EASE co-occurrence scoring system runs at low stringency (GO level 2), although the benefit is a ranking of hundreds of categories associated with a gene list to those central functions passing by Fisher exact test relative to a background file. Central functions passing for >5 genes per category and at an EASE score of P ≤ 0.05 are listed in Table 2. Analysis of the highest-ranking biological themes produced a general categorical association dealing with regulation and control of cell shape and motility, protein degradation and translocation, and cellular energy metabolism. Response signatures could be found for some features within the level-4 heatmaps shown in Figure 6. For example, Functional Group 6 in set k4 consisted entirely of glycolysis genes that would be predicted to have increased during alcohol exposure in the rank order B6N > B6J. Functional Groups 1–2 in set k1 consisted mainly of genes for chaperones and proteasome subunits, respectively, that would be predicted to have decreased during alcohol exposure in the rank order B6N > B6J.

Table 2. Highest-Ranking Biological Themes in the Alcohol Response Based on the Population of Arrayed Genesa
System (Gene Category)LISTPOPULATIONEASE score (P)
  • a

    Gene records (2906) converted to Mmu UniGene. Gene Ontology (GO) annotation analysis performed based on EASE score (P ≤ 0.05); background file contained all arrayed genes merged from the PE+AF array platforms.

Molecular function     
 Actin binding281,0971238,1770.006
 Cytoskeletal protein binding371,0971788,1770.007
 Structural constituent of muscle91,097238,1770.008
 Calmodulin binding151,097558,1770.012
 Ligand-gated ion channel activity161,097668,1770.026
 GABA-A receptor activity81,097238,1770.027
Cellular component     
 Muscle fiber181,049417,840<0.001
 Proteasome complex91,049237,8400.008
 Actin cytoskeleton351,0491757,8400.015
Biological process     
 Main pathways of carbohydrate metabolism211,035687,824<0.001
 Muscle development211,035667,824<0.001
 Cell motility391,0351687,824<0.001
 Muscle contraction181,035587,8240.001
 Glucose metabolism171,035597,8240.003
 Coenzyme and prosthetic group metabolism221,0351047,8240.003
 Energy pathways261,0351117,8240.005
 Tricarboxylic acid cycle81,035187,8240.006
 Alcohol catabolism131,035427,8240.007
 Vesicle-mediated transport331,0351567,8240.007
 Protein complex assembly91,035247,8240.009
 Alcohol metabolism311,0351477,8240.009
 Wnt receptor signaling pathway151,035567,8240.012
 Translational elongation71,035167,8240.013
 Blood coagulation131,035517,8240.031
 Nucleocytoplasmic transport131,035517,8240.031

Statistical Prioritization of Sensitive Pathways

For an overview of metabolic and regulatory pathways affected by alcohol, we matched the 2,906 filtered genes to the pathway database of the KEGG (Kyoto Encyclopedia of Genes and Genomes). Presently, this library has 307 reference pathways ( This analysis returned a representation of differential effects on 10 pathways. To stratify the effect, each pathway was assigned a qualitative “score” on the basis of the number of component genes counted in an expression cluster (Table 3). Scaling the pathway score to the total gene counts in the statistical list produced stratification profiles that directly compared exposure to alcohol between B6J and B6N headfolds (Fig. 8). Using this qualitative scoring method, two pathways (ribosome, proteasome) were predicted for down-regulation in B6N embryos during maternal alcohol intoxication. Two linked pathways (glycolysis-gluconeogenesis, pentose phosphate pathway) were predicted for up-regulation during maternal alcohol intoxication in the rank order B6N > B6J. This trend was validated by expression PCR assay of several glycolysis transcripts as a low-amplitude effect (Fig. 7). Four interconnected pathways involved in the control of cell shape and motility were predicted for up-regulation during maternal alcohol intoxication. These pathways, namely regulation of the actin cytoskeleton, focal adhesion, tight junction, and adherens junction, captured an equal representation of gene records in both substrains (Fig. 8). Two regulatory pathways that are linked to the focal adhesion pathway (Wnt signaling, apoptosis) were predicted for up-regulation in both substrains as well.

Table 3. Highest-Ranking Pathways in the Alcohol Response Based on the KEGG Librarya
KEGG pathwayTermUniGene P valueGenBank totalStatistical listDistribution by clusterScore
  • a

    Significant gene records (2,906) converted to total unique Mmu UniGene records in DAVID 2.1 beta ( for mouse (2,880) as of June 7, 2006. KEGG terms with asterisk (*) also identified when genes were converted to Hsa UniGene records. GenBank total indicates counts in each KEGG pathway versus statistical list (2,906). Distribution by cluster refers to the number of genes belonging to each set from k-means clustering (Fig. 6). With respect to the alcohol response (1 × 2.9 g/kg), gene membership to sets k2 and k5 implies B6J > B6N up-regulation; membership to set k3 implies B6J > B6N down-regulation; membership to sets k4 and k6 implies B6N > B6J up-regulation; and membership to set k1 implies B6N > B6J down-regulation. To compare the substrains, a qualitative pathway “score” is computed based on B6J = (k2 + k5) − k3 and B6N = (k4 + k5) − k1.

MMU04510Focal adhesion*0.0021905512691558+9+12
MMU00030Pentose phosphate pathway*0.002259010530+1+8
MMU04520Adherens junction*0.00574286501043+11+11
MMU04530Tight junction*0.017117316301255+9+12
MMU04310Wnt signaling pathway0.054143356521336+9+10
MMU04810Regulation of actin cytoskeleton*0.0611994594611411+7+4
Figure 8.

Hierarchy of significant KEGG pathways that define the alcohol signature response in B6J and B6N embryos. Pathway score derived from the number of genes in each expression cluster (Table 3) is divided by the number in the significant list genes for that pathway. Values above and below zero reflect pathway up-regulation and down-regulation, respectively, for the response of B6J and B6N cranial neural folds to maternal alcohol treatment (3.0 hr after 1× 2.9 g/kg on day 8).

Figure 7.

Gene expression patterns for glycolysis during maternal alcohol treatment. Microarray data mapped to the Glycolysis and Gluconeogenesis pathway in GenMAPPv2.0 ( Shaded nodes represent genes on the arrays and colored nodes represent genes meeting a 0.8-fold change threshold for up-regulation (red) or down-regulation (green). Low-amplitude up-regulation was evident in the B6N alcohol embryos (A) but not the B6J alcohol embryos (B). This finding was confirmed by global measure of selected genes (PFKL, ALDA, PGK1, PKM2) assayed by expression PCR of independent biological replicates (C). Each group of cranial neural folds was pooled from 1–2 litters to yield n = 5–10 true replicates. PCR signals of glycolysis genes relative to ACTB were merged and normalized to the corresponding saline (control) group. The low-amplitude response of B6N is again evident.

Because gene records compete with one another for membership to each expression group during conventional K-means clustering, the simple scoring of pathways by significant gene counts implies that different subpathways were captured in B6J or B6N embryos. A comparison of input systems into the KEGG focal adhesion pathway, for example, suggests the systems-level responses to alcohol may follow a different genetic program in B6J and B6N embryos. Genes for ECM-receptor interaction and cytokine–cytokine receptor interaction in this pathway showed a general tendency to change in the rank order B6J > B6N. This trend reflected up-regulation in B6J, down-regulation in B6N alcohol-treated embryos, or both (Table 4). For a more comprehensive analysis of genetic differences, we selected significant genes associated with all 10 affected KEGG pathways (Fig. 8). These genes were then applied to preliminary computational gene network prediction by Pathway Architect v1.2.0 (Stratagene, La Jolla, CA). Gene records were portrayed as an interacting network and the two small molecules (EtOH, PK11195) were added to the representation. Ethanol had several potential entries into the network, including MAPK1, ALDH3A2, CD14, PFKM, TNFRSF1A, RPS6, IGF1, EGFR, and PTEN, whereas, PK11195 primarily entered at AKT1 (Fig. 9).

Table 4. Alcohol Effects on Selected Input Genes in the Focal Adhesion System at 3.0 hr
Systematic nameGeneB6JB6NB6J-B6N difference
  1. aAsterisk (*) denotes significant genes from the 2906 list based upon input systems for KEGG pathway MMU04510 (focal adhesion); all other genes were retrieved from the list by membership to the same functional classification and expression profile in the significant gene list (Fig. 6) or Biocarta EPH receptor pathway ( Arrows indicate up/down regulation in alcohol-treated versus control heaofold at threshold of 0.75-fold change (log2 = ± 0.375); dash (—) means the ratio did not pass threshold. The substrain differential is given in log2 units (e.g. 1 = two-fold change).

BF227507*, Z74516*Procollagen 1 alpha 2+0.77
NM_016919*Procollagen 5 alpha 3−0.53
M34573*Procollagen 6 alpha 2+0.74
BI664675*Integrin alpha 3+1.09
NM_016780*Integrin beta 3−0.35
X51841*Integrin beta 4−0.58
Z26653*Laminin alpha 2+0.96
M55618*, NM_011607*Tenascin C (hexabrachion)−0.34
NM_008478L1 cell adhesion molecule+0.20
AV362920*Focal adhesion kinase (PTK2)+1.28
U09303Ephrin b1+1.38
BM24734Ephrin b2+1.61
U66406Ephrin b3+0.88
AI854630Ephrin a5+2.24
M59371EPH receptor A2+0.68
L36645EPH receptor A4+1.34
BG060788*HGF receptor−0.44
AK014017*EGF receptor+0.22
M58051*FGF receptor 3+0.74
BE960124*IGF-1 receptor+0.70
Figure 9.

Computational gene network prediction based on genes associated with the significant KEGG pathways. Gene records for the significantly altered genes in the derived KEGG pathways were portrayed as an interacting network using PathwayArchitect v1.2.0 (Stratagene). Gene products are coded in red and small molecules applied to this study (EtOH, PK11195) in green. Lines indicate curated interactions in the ResNet database for direct binding (purple), direct regulation (blue), and direct modulation (grey).

As validation, we analyzed independent RNA samples (n = 2–3 per condition) using primers specific for murine AKT1, AKT2, and PTEN (Fig. 10). Basal expression was similar across substrains, and was minimally affected by the TSPO ligands PK11195 (partial antagonist) or Ro5-4864 (agonist) given alone. In B6N embryos, all three transcripts were elevated over the basal level at 3 hr after maternal alcohol treatment. Basal expression was observed when dams were cotreated with alcohol and PK11195, but not with alcohol and Ro5-4864. A tempered response was evident in B6J embryos although a weak trend was evident.

Figure 10.

Expression PCR analysis of Akt1, Akt2, and Pten transcripts in the embryonic headfold. B6J and B6N samples were harvested 3.0 hr post-injection on day 8 of gestation. Total RNA from the microdissected headfold was subjected to one round of amplification (aRNA) followed by RT-PCR (lanes 1–4) and separation on the Agilent 2100 Bioanalyzer. A comparison to PCR reactions of the total (unamplified) RNA pool is shown (lanes 5–8). Lanes: marker (L); Actb (lanes 1 and 5); Akt1 (lanes 2 and 6); Akt2 (lanes 3 and 7); and Pten (lanes 4 and 8). PCR cycles were optimized for linear signal (24 cycles for Actb and 27 cycles for all others). Histograms show mean signal relative to Actb and standard error from independent biological replicates (n = 2–3) per condition. Samples were: sal (normal saline treatment), EtOH (2.9 g/kg alcohol), PK11195 (4 mg/kg), Ro5-4864 (4 mg/kg), CoT (EtOH + PK11195 cotreatment), and CmB (combined effect of EtOH + Ro5-4864).


The results from this study show that developmental effects of maternal alcohol treatment varied in two closely related C57BL/6 genetic lines (B6J, B6N). Similar degrees of FWR occurred in both substrains whereas the risk of malformations clearly differed (B6J > B6N). Counter-exposure with 4 mg/kg PK11195 significantly protected B6J embryos from alcohol-induced malformations but promoted malformations in a subset of B6N embryos. This implies a role for mitochondrial TSPO protein in the differential susceptibility to alcohol. Comparative bioinformatics analysis of gene expression in the embryonic headfold at 3.0 hr after the first maternal alcohol injection revealed metabolic and cellular reprogramming that was substrain-specific and/or PK11195-dependent. Because this dimorphism correlated with the risk of malformations, the genomic response can be anchored with genetic susceptibility. Mapping to KEGG pathways revealed down-regulation of ribosomal proteins and proteasome, and up-regulation of glycolysis and pentose phosphate pathway in B6N embryos; and significant up-regulation of tight junction, focal adhesion, adherens junction, and regulation of the actin cytoskeleton pathways (and a near-significant up-regulation of Wnt signaling and apoptosis pathways) in both substrains.

Risk for Alcohol-Induced Malformations, But Not Fetal Weight Reduction, Differed Between Substrains

Alcohol's effect on B6J pregnancies is well established following a double injection of 2.9 g/kg ethanol spaced 4 hr apart (Webster et al.,1980; Sulik and Johnston,1983; Sulik et al.,1986; Kotch and Sulik,1992; Sulik,2005). Although each injection may drive maternal BAC to high levels, the penetrance of FASD-like malformations is incomplete. In comparing B6J and B6N substrains, we observed the full range of expected outcomes in B6J pregnancies (15% FWR, 34% resorptions, and 35% malformations) but only FWR in B6N pregnancies. Most frequently, the observed malformations involved only the eye. Alcohol treatment raised the incidence of visible eye malformations 5% above background in B6N mice (insignificant trend) and 30% above background in B6J mice (highly significant effect). This confirms the early optic primordium as a critical target of alcohol (Cook et al.,1987) and other teratogens (Wubah et al.,2001) in mouse embryos at the 4–8 somite pair stage. Ocular defects such as coloboma, microphthalmia, persistent hyperplastic primary vitreous body, optic nerve hypoplasia, and funduscopic abnormalities of the optic disk are common clinical findings in FASD (Strömland,1985, 1987; Strömland and Pinazo-Duran,1994). Because our evaluation focused on eye reduction defects (coloboma, microphthalmia, anophthalmia), the full range of ocular malformations, and their possible relationship to disturbances of forebrain development in general, cannot be determined at this time.

Experimental variables accounting for an interaction between treatment and susceptibility are complex and may include advanced maternal age (personal observation), dosing scenario (Gohlke et al.,2005), developmental timing (Ogawa et al.,2005), and genetic variation (Warren and Li,2005). Efforts were undertaken to minimize maternal age and weight variables by purchasing B6J and B6N mice at about the same time. Although dams cohabited a dedicated animal room and received the same diet, B6N dams at conception were slightly heavier than B6J dams (23.8 g vs. 21.6 g, respectively). Since we do not know if these murine lines follow the same age-weight trajectory, we cannot rule out the possibility that slight maternal age or weight differences might contribute to the differential susceptibility to alcohol. Nor do we know the potential contribution of pharmacokinetic differences in alcohol adsorption, distribution, metabolism, and excretion.

Parenteric (i.p.) dosing with alcohol as applied here circumvents intestinal absorption and thus may lead to high blood alcohol content (BAC) transients. For example, a 25-g pregnant mouse of lean body mass (77% body weight composition by water) dosed with 2.9 g/kg alcohol achieves a whole-body concentration of 72.5 mg alcohol per 19.3 ml body water. The instantaneous theoretical peak of 3.76 mg/ml (376 mg%) BAC is consistent with empirical measures (Webster et al.,1980; Maier et al.,1999) but over 4-fold higher than the legal limit for human consumption in most U.S. states. While BACs of this magnitude are achieved by individuals who have acquired alcohol tolerance, the parenteric exposure scenario has limitations with respect to general applicability. For this reason, oral dosing (Chernoff,1980; Boehm et al.,1997) or self-administration (Parnell et al.,2006) scenarios have been applied to more closely reflect the human condition. Future studies should be conducted to determine whether the rank order sensitivity B6J > B6N is followed during enteral exposure scenarios. On the other hand, the differential effects in B6J and B6N substrains partially overcome the limitation of parenteric dosing insofar as the latter embryos can withstand heavy binge-like exposure and, hence, possess genetic and/or epigenetic variations of potential medical importance.

FASD studies in hybrid murine strains have almost invariably linked the differential susceptibility to alcohol-induced malformations, but not FWR, to a maternal or maternal-fetal interactive effect (Chernoff,1980; Gilliam and Irtenkauf,1990; Gilliam et al.,1997; Boehem et al.,1997). Because FWR was similar between B6J and B6N substrains, our findings support the notion that different genetic/epigenetic factors may account for alcohol-induced malformations primarily. Whereas the present study is the first to compare these substrains for prenatal alcohol exposure, we cannot rule out the possibility that substrain differences in the pharmacokinetics of alcohol or the developmental rate of the embryos contributed to the observed differential susceptibility between substrains. B6N and B6J mice, although closely related genetically, are not identical. Of 867 tested microsatellite markers spanning 19 chromosomes, 1.6% were polymorphic between these substrains (Hovland et al.,2000). These substrains are known to differ in various physiological parameters, including relative AhR affinity (He et al.,1997), susceptibility to cadmium teratogenicity (B6N > B6J) (Hovland et al.,2000), susceptibility to chemical-induced colon tumors (B6N > B6J) (Diwan and Blackman,1980) and cardiac dysfunction during ketamine-midazolam anesthesia (Roth et al.,2002).

Pharmacological Intervention for FASD at the Mitochondrial Recognition Site for PK11195

The 18-kDa translocator protein (TSPO) (Papadopoulos et al.,2006), formerly known as the peripheral benzodiazepine receptor (Bzrp, PBR), is a highly conserved five-pass protein that has close structural and functional similarity to the tryptophan-rich sensory protein (TspO) in the oxygen-sensing proteobacteria, Rhodobacter sphaeroides (Yeliseev et al.,1997; Yeliseev and Kaplan,2000). Whereas TSPO predominantly localizes to the mammalian outer mitochondrial membrane (Lacapère and Papadopoulos,2003), TspO localizes to the outer protobacterial membrane and serves as an oxygen-dependent signal generator that negatively regulates photosynthetic gene expression when the bacterial environment favors aerobic growth (Yeliseev et al.,1997). General functions of mammalian TSPO include mitochondrial cholesterol transport (Li et al.,2001), oxygen-sensing (O'Hara et al.,2003), and neuroprotection (Ryu et al.,2005). TSPO is the molecular target for PK11195. Tissue binding sites for PK11195 increase in adult brain during alcohol exposure (Schoemaker et al1983; Obernier et al.,2002); hence, TSPO may function directly or indirectly in alcohol's mode of action.

Treating B6J dams with 4 mg/kg PK11195 rescued B6J fetuses from alcohol-induced malformations, an observation consistent with teratological findings in outbred (CD-1) mouse embryos exposed to methylmercury (O'Hara et al.,2002), 2-chloro-2′-deoxyadenosine (Charlap et al.,2003), or transitive hypoxia (O'Hara et al.,2003). In contrast, B6N litters showed severe fetal malformations, albeit at low incidence, in those dams co-treated with alcohol + PK11195. Further research is required to more accurately assess the potential role of TSPO in the mode of action of alcohol and its utility as a druggable target for the treatment of FASD. It is, however, curious how PK11195 might sensitize B6N embryos at low-risk to alcohol-induced damage while substantially rescuing B6J embryos at high-risk for FASD malformations. Perhaps the mechanism relates to local oxygen metabolism or the generation of reactive oxygen species (ROS). ROS induces polymerization of TSPO monomers to higher molecular weight forms that have an increased capacity for cholesterol binding (Delavoie et al.,2003). Such a change during alcohol exposure (B6J) might over-ride a homeostatic function of TSPO (B6N) with either process perhaps sensitive to PK11195. Our unifying hypothesis thus considers TSPO a repressive “oxygen-sensing cholesterol capacitor” in the embryo that couples cholesterol transport with oxygen metabolism (Knudsen and Green,2004).

TSPO-Sensitive Pathways and Gene Networks During the Initiation of Ethanol-Induced Malformations

In the search for ethanol-sensitive genes, both a priori and non a priori approaches have been taken in an effort to characterize developmentally important responses (Lee et al.,1997; Rifas et al.,1997; Ahlgren et al.,2002; Poggi et al.,2003; da Lee et al.,2004; Xu et al.,2005; Yamada et al.,2005). The present study used a non a priori approach in which essentially all transcripts were assayed by DNA microarray. Using various statistical (ANOVA) and multivariate approaches, we folded the genomic response of the headfold 3.0 hr post-treatment into genes discriminating statistically between low- (B6N) and high- (B6J) risk conditions. High blood alcohol content must be sustained in dams for several hours to invoke teratogenesis, and this is traditionally accomplished by a double injection 4 hr apart. Transcript profiling revealed changes already at 3 hr after the first injection that discriminated the substrains with respect to their risk to alcohol-related birth defects. Since these embryos were harvested for genetic analysis 1 hr before the dams would have gotten the second alcohol injection, we can establish a pattern in the differential risk of B6J and B6N embryos during the critical exposure period, prior to the second maternal injection that invokes greater teratogenicity. This dose differentiated significant substrain-specific and PK11195-dependent responses of the headfold.

By coupling statistical methods with bioinformatics programming, we were able to prioritize genes and pathways as well as candidate genes from the literature with regard to the relative risk of alcohol-related birth defects. Results for some candidate genes (Msx2, Gli1, plunc) did not fit a priori patterns drawn from the literature (Rifas et al.,1997; Ahlgren et al.,2002; da Lee et al.,2004; Yamada et al.,2005). We do not know whether the disparity reflects a limitation in our study design or model; however, our screen returned a number of genes with patterns that fit a priori predictions. The GABAA receptor, for example, showed changes consistent with studies indicating these receptors mediate developmental and behavioral effects of alcohol (Toso et al.,2006, Hanchar et al.,2006). We also found changes in genes for inositol-phosphate 3-kinase (IP3K) signaling as reported (Hard et al.,2005; Kerns et al.,2005). This included Akt1 and Pten expression. Thus, up-regulation in the headfold was validated by independent PCR methods as a low-risk effect (B6N > B6J). This fits the a priori expectation because down-regulation of Akt1 (Hard et al.,2005) or Pten (Kerns et al.,2005) has been reported in systems sensitive to alcohol exposure.

For non a priori analysis, we focused on the curated KEGG pathways. This revealed two general trends in the headfold. The first pertained to pathway genes affected in one substrain only: two were down-regulated (ribosomal proteins, proteasome) and two up-regulated (glycolysis-gluconeogenesis, pentose phosphate pathway) specifically in the B6N headfold. At least the former response is consistent with other systems (Lang et al.,2000; Rahman and Miles,2001; Gutala et al.,2004) and the latter was validated by independent PCR as a low-amplitude response. The second general trend in the data pertained to KEGG pathways altered in both substrains: tight junction, focal adhesion, and adherens junction pathways were the most significant, and regulation of the actin cytoskeleton, Wnt signaling, and programmed cell death (apoptosis) appeared at lower confidence. Our screen detected changes in some cytoskeletal genes (e.g., cytokeratins, tubulins, tropomyosin) described in the FAS/FASD literature. For example, non-muscle alpha-tropomyosin (Tpm2) was identified as an alcohol-inducible transcript by differential display analysis (Lee et al.,1997). Those investigators concluded that ethanol must “alter microfilament function during critical periods of embryogenesis and affect cell morphology and migration of early neuronal cells”; however, that screen was performed in the B6N embryo whereby the risk to FAS may be low. Alcohol promoted tubulin gene and protein expression in other systems to affect intracellular trafficking and ultimately cell size and transport functions (Azorin et al.,2004). Consistent with these findings, our microarray-based screen detected significant up-regulation of Tpm2 as well as several tropomyosin-related genes, tubulin-beta (Tubb2a) and other transcripts in regulation of the cytoskeleton (B6J > B6N). Many genes in cytoskeletal pathways are targets for regulatory signals that may also govern growth-realted processes. For example, the focal adhesion pathway directly projects to the Wnt signaling and apoptosis pathways that are important for lens and retinal development (Kubo et al.,2005; Fokina and Frolova2006). Taken together, our findings are consistent with the emerging view that alcohol's developmental effects are mediated by common signaling pathways linking receptor activation to cytoskeletal reorganization (Lindsley et al.,2006). We further suggest that these responses may be connected with signals integrated by TSPO, the mitochondrial recognition site for PK11195.

There is functional evidence for detrimental effects of alcohol on cell–cell adhesion molecules, especially the L1 cell adhesion molecule (L1 CAM) (Charness et al.,1994; Bearer et al.,1999; Chen et al.,2005; Lindsley et al.,2006). Our findings are consistent with a priori analysis in FASD (Miller et al.,2006) that focused on transcripts for several integrins and L1 CAM. Through comparative bioinformatics, however, the non a priori screen implies a broader involvement of the focal adhesion complex. This receptor-mediated cell adhesion system includes integrins, L1 CAM, and ephrin ligands that comprise juxtacrine signaling inputs, and growth factor receptor tyrosine kinases (RTKs) that comprise paracrine signaling inputs. Functional activation of the focal adhesion complex phosphorylates focal adhesion kinase p125FAK in the cytosol to promote cell adhesion; in contrast, the release of cells from the focal adhesion complex facilitates cell migration. Thus, we may speculate that a net increase in expression of juxtacrine over paracrine signaling inputs during alcohol exposure drives the state of the system toward cell adhesion (B6J) or cell migration (B6N). This may serve to compensate for detrimental effects of alcohol on cell–cell adhesion molecules in B6J embryos, or alternatively to reprogram cellular behavior in B6N embryos.

Resolving Genetic Networks Associated With a Metastable State

The differential reprogramming of diverse metabolic and regulatory pathways in the headfold identified through microarray profiling raises questions as to how genes work together as co-expression networks, and the critical rate-limiting steps that control the network. Dynamic systems are said to be stable if they recover to the original state following disturbance (homeostasis), but unstable if they continue to move away from their original state. If we consider the headfold as a system in metastable equilibrium, then its ability to withstand alcohol-induced stress may ultimately depend on the robustness of gene networks as well as the amount and duration of exposure. Knowing the key informational hubs in these networks can enable new developmental biomarkers as metrics to quantitatively assess the physiological state of the embryo and how developing systems behave when challenged with ethanol. Consider the focal adhesion complex, for example. One potential mechanism that may integrate common signaling pathways linking receptor activation to cytoskeletal reorganization is the insulin signaling pathway. Expression networks constructed computationally from genes in the altered KEGG pathways identified entry points for ethanol at several hubs (MAPK1, ALDH3A2, CD14, PFKM, TNFRSF1A, RPS6, IGF1, EGFR, PTEN) and for PK11195 at AKT1. AKT1/protein kinase B is a serine-threonine protein kinase rapidly activated by insulin and other growth factors through IP3K and deactivated by phosphatase tensin homologue (PTEN). Reduced bioavailability of IGF-1 has been associated with decreased embryo weight in alcohol-treated chick embryos (Lynch et al.,2001). However, the increase in AKT1 and PTEN observed in the present study may reflect robustness of a gene network in B6N embryos.

In conclusion, our results are consistent with the emerging view that alcohol disrupts common signaling pathways linking receptor activation to cytoskeletal reorganization (Lindsley et al.,2006). Our findings further suggest that critical signals are integrated with the mitochondrial recognition site for PK11195 in the mode of action of alcohol. Regulation of the embryonic transcriptome holds important information about programmed responses to restore homeostasis and can reveal when a weak or incomplete adaptive response may invoke general systems failure.



Standard biochemicals were purchased from Fisher Scientific (Fairlawn, NJ) or Sigma Chemical Company (St. Louis, MO) and were of the highest grade available. 1-(2-Chlorophenyl)-N-methyl-N-(1-methylpropyl)-3-isoquinolinecarboxamide (PK11195) and the ligand-solubilizer 2-hydroxypropyl-β-cyclodextrin (HP-beta-CD) were from Sigma-RBI. RNeasy micro-spin kits were from QIAGEN, Inc., (Valencia, CA). Superscript II RT, random primers (HAS701), RNase-free DNase I, and other reverse-transcriptase PCR (RT-PCR) reagents were from Invitrogen (Chicago, IL). Ready-to-go PCR beads were from Amersham (Piscataway, NJ). Water for solutions and buffers was collected at 18.2 megaohms-cm from the Milli-Q Plus Ultra-pure water system from the Millipore Corporation (Milford, MA).


Animal protocols complied with the Care and Use of Laboratory Animals as approved by the Institutional Animal Care Use and Committees of the performance sites for this study, Thomas Jefferson University and the University of Louisville. The study used inbred C57BL/6NCrl (B6N) mice purchased from Charles River Laboratories (Wilmington, MA), and C57BL/6J (B6J) mice purchased from the Jackson Laboratory (Bar Harbor, ME). Some experiments also used outbred CD-1 mice from Charles River. The two lines of C57BL/6 mice represent “substrains” bred at The Jackson Laboratories (J), the NIH and more recently Charles River Laboratories (NCrl). Mice were purchased at approximately 18–19 g and acclimated for 2 weeks on a 12-hr photoperiod (06.00–18.00 hr light) before breeding. The B6N and B6J colonies were housed in the same room. Males were caged individually and females five or less in static microisolation cages with bedding that is absorbent, non-nutritive, and non-toxic. Diet was Purina mouse chow and tap water ad libitum.

Alcohol and PK11195 Treatments

To generate timed pregnancies, males were random-bred to nulliparous females of the same substrain (21–23 g body weight on average) at 07.30–8.00 hr. Detection of a vaginal plug at 13.30–14.00 hr was regarded as evidence of coitus (day 0). Dams were considered pregnant if they showed 2–3 g weight gain at 09.30 hr on day 8. All treatments were performed at that stage by intraperitoneal (i.p.) injection of test solution at 0.5 ml per 30 g maternal body weight. Control dams received vehicle (saline) or were untreated as indicated. For alcohol exposure, the injection was 22% absolute ethanol (v/v) in isotonic saline (2.9 g ethanol per kg body weight). For phenotype evaluation, we used the well-described model of two i.p. injections spaced 4.0 hr apart (Webster et al.,1983; Kotch and Sulik,1992). For gene expression analysis, only the first injection was given. PK11195 was prepared in 4.5% HP-beta-CD (1:11 molar ratio of drug:exclusion compound) in isotonic saline and injected i.p. to dose at 4.0 mg/kg PK11195 (Charlap et al.,2003; O'Hara et al.,2002), coincident with ethanol injection.

Sample Collection

Treated dams were euthanized by carbon dioxide asphyxiation on days 8, 11, 14, or 17 of gestation. RNA analysis was limited to day 8 embryos having 4–8 somite pairs at the time of procurement. Embryos were rinsed in ice-cold Hank's Balanced Saline Solution (HBSS) and the headfold was microdissected by a transverse cut immediately rostral to the heart. Cell types in the sample included neural and surface ectoderm, cranial neural crest, mesoderm, and foregut endoderm. Due to the restriction on somite-stage and the fact that not all embryos met the stage criteria, sufficient tissue for RNA analysis required pooling samples from 1–2 litters. For preliminary morphology, several litters were harvested on gestational day 11 and embryos were processed for scanning electron microscopy (Sulik et al.,1986; Kotch and Sulik,1992), or on day 14 for routine paraffin sectioning and hematoxylin-eosin staining. For teratological evaluation, the litters were autopsied on day 17.

Teratological Evaluation

The uterus was examined for resorptions and fetuses were weighed and fixed (after hypothermia) in neutral-buffered formalin for evaluation. Each fetus was rinsed in water and phosphate-buffered saline (PBS) and examined under a stereoscope. Ocular phenotypes were scored for each eye as follows: 0 = normophthalmia, 1 = irregular iris/coloboma or narrowed pupillary ring, 2 = visible small eye, 3 = very small eye or severe microphthalmia, and 4 = no eye visible (complete apparent anophthalmia). An “eye reduction score” (ERD) was calculated by tabulating the total score and dividing by the number of eyes examined (Wubah et al.,2001). After recording other external defects, several gross dimensions were measured using an electronic digital micrometer (Marathon CO 030025): crown-rump length (CRL), naso-occipital (NO), and trans-temporal (TT) dimensions. Cranial circumference (CRC) was then estimated by the formula CRC = (2/3NO+TT)/2 * TT (O'Hara et al.,2002). The litter was used as the sampling unit for statistical analysis. Two-way ANOVA (background, treatment) was run with GraphPad Prism version 4.02 for Windows, GraphPad Software, San Diego California, When two-way ANOVA revealed a significant group-wise effect (P ≤ 0.05), post-analysis was performed using Bonferroni-corrected multiple comparison tests, as indicated.

Microarray Experiment Design

B6N and B6J embryos were procured 3.0 hr after maternal injection with normal saline, 2.9 g/kg ethanol in saline, or the same dose of ethanol + 4.0 mg/kg PK11195 in HP-beta-CD. Headfold samples were analyzed in essentially the same experiment repeated independently on two different microarray platforms (Perkin Elmer [PE] and Affymetrix [AF]). The resulting datasets were submitted to the Gene Expression Omnibus ( respository with the series title “Strain-dependent effects of alcohol on early mouse embryos” (GSE1074). The first dataset (PE), run April 2002 to January 2003, used samples pooled from 2 litters and the platform was a two-channel MPS621 array ( This platform, now retired, has 4,800 sequence-verified gene elements derived from over 50 different human cDNA libraries reflecting a variety of well-annotated cellular processes and disease pathways. The second dataset (AF), run May 2005 to November 2005, used samples pooled from 1 litter and the platform was Affymetrix GeneChip® Mouse Genome 430A 2.0 Array ( The MG430A 2.0 platform has 45,102 oligonucleotide probe cells representing approximately 14,000 well-characterized genes in the draft mouse genome assembly. The design matrix thus consisted of 2 substrains (B6J, B6N) and 3 exposures (saline, alcohol, cotreated), and it represented 4 biological replicates across two platforms (PE1, PE2, AF3, AF4).

RNA Isolation

Total cellular RNA was isolated as described (Nemeth et al.,2005). Sufficient samples were pooled to yield 15–30 μg total cellular RNA. Tissues were homogenized in Lysis Buffer RLT and an equal volume of 70% ethanol was added. Samples were extracted with RNeasy (Qiagen) spin columns and centrifugation at 10,000 rpm. This method removes small RNAs such as the 5S and transfer RNA bands, but leaves RNA molecules longer than 200 nucleotides. The RNA was stored at −80°C until used. Purity and intactness were assessed with RNA 6000 Nano-LabChip separation on the Agilent 2100 Bioanalyzer and Nanodrop ND-100 (Nanodrop Technologies, Wilmington, DE) utilizing A260/A280. A 2:1 ratio of 28S/18S rRNA was the acceptance criterion. All RNA samples were subjected to a second quality-assurance step using semi-quantitative reverse transcription-polymerase chain reactions (RT-PCR) with positive control genes (β-actin, 16S rRNA). Where indicated, headfold RNA (0.5 μg) was further amplified using Agilent Low RNA Input Fluorescent Linear Amplification Kit. The fluorescent cRNA Synthesis protocol was followed, excluding the addition of cyanine dyes (Agilent Technologies). First round gave 1,000× quantitative amplification of low- and high-abundance transcripts and no degradation of signal quality.

RNA Labeling and Hybridization

PE arrays.

RNA was converted to cDNA with incorporation of biotin-11-dCTP (BN) or fluorescein-12-dCTP (FL) for two-channel hybridization. The RT step used poly(dT)-anchored priming; hence, labeled target cDNA species would have bias of the 3′-end of the transcript. Label efficiency was determined by the “3rd dot rule” in a serial dilution of labeled sample versus pre-labeled standard using a quantitative immunoblot assay (Knudsen et al.,2003). Each array was replicated from independent samples using the reverse label assignments (Nemeth et al.,2005). Labeled target RNA (15–20 ng) was hybridized at 56–59°C competitively to MPS621 probe. Post-hybridization conditions were 0.5× SSC/0.1% SDS for 10 min followed by 0.06× SSC/0.01% SDS and 0.06× SSC.

PE integrative chemistry culminates in signal amplification and greatly enhanced detection using two-channel indirect labeling of BN- and FL-labeled target RNA with post-hybridization amplification through conventional immunodetection and tyramide signal amplification (TSA) (Alder et al.,2000). TSA used peroxidase-conjugated anti-FL and cyanine-3-tyramide (Cy3) followed by peroxidase-conjugated streptavidin and cyanine-5-tyramide (Cy5). Slides were washed with 0.1 M Tris buffer, pH 7.6, containing 0.15 M NaCl and 0.05% Tween 20, then 0.06× SSC and spin-dried. Fluorescent images of Cy3 and Cy5 channels were generated by ScanArray 5000 XL with spot-finding software, using median signal with local background correction.

AF arrays.

RNA was subjected to two-cycle amplification and labeling. The first cycle reverse transcribed the poly(A)-rich RNA to cDNA. First-strand synthesis used T7-Oligo(dT) Promoter Primer (50 μM) and Superscript II under RNase-free conditions, in parallel with a control template for spiking the reaction. The first-round mix was incubated 1 hr at 42°C, deactivated 6 min at 70°C, and stored at −20°C. Second-strand synthesis was mediated by DNA polymerase I and RNase H for 2 hr at 16°C followed by deactivation for 10 min at 75°C. Double-stranded cDNA was purified as a template for in vitro transcription (IVT) for complementary RNA (cRNA) amplification. The IVT reaction was carried out with T7 RNA Polymerase and unlabeled ribonucleotides for 16 hr at 37°C. After clean-up, the cRNA was subjected to QA/QC assessment as above, for a yield of ∼600 ng. The second cycle reverse transcribed the unlabeled cRNA to cDNA. First-strand synthesis used random primers (3 μg/μl) and Superscript II, 1 hr at 42°C. The second strand was synthesized with T7-Oligo(dT) Promoter Primer, DNA polymerase I, and RNase H to generate a double-stranded cDNA template containing T7 promoter sequences. Cleanup of double-stranded cDNA for the target labeling assays was performed with the GeneChip cDNA Cleanup kit. The template was further amplified with T4 DNA polymerase and labeling with biotinylated nucleotide analog/ribonucleotide mix in the second IVT reaction (16 hr, 37°C). Biotinylated cRNA was finally cleaned up and fragmented by limited hydrolysis to a distribution of RNA fragment sizes below 200 bases. The target cRNA was subjected to a final QA/QC and stored at −80°C.

Hybridization mix consisted of Control Oligo B2, Eukaryotic Hybridization Buffer, Herring Sperm, BSA, and DMSO added to 15 μg fragmented target cRNA. After denaturing, the mix was hybridized (16 hr, 45°C) to Affymetrix GeneChip® Mouse Genome 430A 2.0 Array. All chips were from the same manufactured lot. Arrays were imaged with the GeneChip® 3000 scanner. Data from the 12 *.cel files were normalized using RMA ( Values recorded here reflect intensity (I) derived from the scanner.

Data Normalization and Statistics for Differential Expression

PE arrays.

The two-channel PE series competitively hybridized test and reference samples to yield quotients for the effect relationship of alcohol-to-saline and cotreated-to-alcohol. Expression ratios were determined from background-corrected signal in alcohol/saline and cotreated/alcohol hybridizations. The expression ratio, T, was determined for microarray data by computing the background-corrected fluorescent signal from the query sample (Q)/reference sample (R). Ratiometric data were transformed to log2-space to produce a continuous spectrum of up- and down-regulated values. Data were normalized by plotting the difference, log2(Q/R), against the average, (1/2)log2(Q*R) followed by the application of locally weighted regression (lowess) to smooth intensity-dependent ratios (Yang et al.,2001). Paired-slides normalization was applied to the replica swaps to force Cy3/Cy5 = Cy5/Cy3 for each gene element across all conditions (Nemeth et al.,2005). We then polished each array to MEDIAN = 0.00 and scaled to STDEV = 0.50 (Singh et al.,2005). The normalized data were fitted to a linear model for microarray analysis using the “limma” package ( This algorithm computes a coefficient for each element and sample pair average across 4 dimensions and 2,382 gene elements (average of duplicate features on the array, less negative-control spots). The return is a measure of the M-value (M) as the log2 fold-change for that gene derived from the “T” ratio, and the A-value (A) as the average expression ratio for that gene across the 2382-probe series. The P value was obtained from the moderated t-statistic (eBayes function in R) after applying the Benjamini-Hochberg false discovery rate correction to protect against false positives in the multiple testing scenario. Of 2,382, the number of probes passing the initial statistical filter was 628 (P ≤ 0.005).

AF arrays.

Probe-level data normalization used the robust multichip average (RMA). Log2-transformed datasets were subjected to four-way ANOVA (treatment, strain, labeling, fluidics batch) using Partek Genomics Solution Software ( The Sources of Variation plot indicated the signal-to-noise values across the 12 AF chips were very high for biological effects (values = 11.3 and 49.5 for treatment and strain, respectively) and very low for procedural effects (0.81 and 0.18 for labeling and fluidics batch, respectively). Data were analyzed by fitting to the linear model and computing the coefficients for each element and sample pair average across 6 dimensions and 45,037 probe features (less control elements). In this case, the M-value refers to the log2 expression level for that gene and the A-value is the average expression level for that gene across the AF series. The comparisons of biological interest were selected to identify genes altered by treatments in B6J embryos; genes altered by treatments in B6N embryos; and genes differently expressed in B6N compared to B6J embryos irrespective of treatment. The first and second parameters had three contrasts each and the third parameter had four contrasts from the interaction terms. Summary statistics were computed by the eBayes function in R to calculate the P value from the moderated t-statistic for each contrast with Benjamini-Hochberg correction for multiple testing. Of 45,037 features, the number of probes passing the initial statistical filter was 2340 (P ≤ 0.005). Expression ratios were computed for alcohol/saline and cotreated/alcohol post-hoc by taking the average saline and alcohol values from corresponding samples as the reference denominator.

Data Merging and Clustering

Co-inertia analysis (CIA) is a multivariate method of data coupling that identifies trends or patterns as co-relationships in multiple studies containing the same (or highly related) samples. It can be used for overlaying microarray data from different gene expression platforms for highly related samples because it is independent of probe annotation (Culhane et al.,2003). The package ADE-4 was downloaded from the R statistical computing software home page ( CIA was applied to overlay replicate alcohol/saline and cotreated/alcohol samples (B6N, B6J) from the PE and AF datasets and to finally compare the intrinsic data structure between platforms. We converted the AF probe IDs to GenBank accession numbers and the normalized AF+PE data were finally merged into a hybrid dataset that had 2,906 non-redundant genes. These genes were partitioned into 6 expression profiles using K-means clustering and Pearson correlation (GeneSpring v 7.2).

Functional Annotation

We used the NIH/NIAID Database for Annotation, Visualization and Integrated Discovery (DAVID) (Dennis et al.,2003). This evidence-based method groups tens of million of gene/protein identifiers from over 65,000 species into 1.5 million unique records and then ranks functional categories in the list based on co-occurrence by number of genes enriched within each category, or by statistical over-representation relative to the proteome of a given species. Because the statistical gene list was merged from murine (AF) and human (PE) platforms, we converted GenBank records into UniGene identifiers with BDSM ( to facilitate functional annotation.

The Functional Annotation tool of DAVID 2.1 beta ( was used as a scoring system to classify genes by gene-term association. To compensate for annotation amplification or attenuation during cross-species conversion, we ran the analysis on each orthologous list. The highest-ranking biological themes were stratified by Gene Ontology (GO) terms. This analysis was run at high stringency (level 4) and P ≤ 0.005, capturing those terms above a lower threshold = 4 genes and eliminating the most general terms above an upper limit = 80 genes. Clusters were also analyzed by the EASEonline module ( to derive the highest-ranking biological themes based on the population of arrayed genes (P ≤ 0.05) (Hosack et al.,2003). The background file was built from UniGene records derived from the AF+PE probe sets.

Expression PCR

RT-PCR with appropriate gene-specific primers was used for assessing relative transcript abundance on independent (non-arrayed) samples. RNA (1 to 5 μg) passing the acceptance criteria was reverse transcribed with Superscript II RT at 42°C in the presence of dithiothreitol, random primers, and 0.2 mM each dNTP. Negative controls replaced reverse transcriptase with an equal volume of DEPC-treated water. PCR reactions used Ready-to-go PCR beads, 2 μl of each sample cDNA, and 5 μl each primer in 25 μl volume completed with 18-mΩ water. PCR cycles were 95°C (1 min), 66°C (2 min), and 72°C (1 min). Specific primers were designed from murine GenBank sequences using OLIGO Primer Analysis Software (version 5.0, National Biosciences, Inc.). Primer sequences and PCR product sizes are listed in Table 5. Preliminary runs were conducted to optimize cycle numbers (24–33) for the linear range. Generally this meant 24 or 27 cycles for β-actin and 27 or 30 cycles for the target genes. PCR products were analyzed using 8% polyacrylamide gel electrophoresis, ethidium bromide staining, and imaging software; or microfluidics-based platform on the Agilent 2100 Bioanalyzer with DNA 1000 LabChip. Results for specific target gene signals were referenced to the β-actin signal. One-way analysis of variance (ANOVA) with multiple comparison post-testing was performed using GraphPad Prism version 4.02.

Table 5. Mouse PCR Primers Used in This Study
Gene symbolGene productRefSeqUpper primer (5′ → 3′)Lower primer (5′ → 3′)bp PCR product
ActbActin, beta, cytoplasmicNM_007393.1TAC CAC AGG CAT TGT GAT GGAAT AGT GAT GAC CTG GCC GT310
Akt1Thymoma viral proto-oncogene 1NM_009652ATG AGG TTG CCC ACA CGC TTA CCAC AGC CCG AAG TCC GTT ATC319
Akt2Thymoma viral proto-oncogene 2NM_007434GAA GAC TGA GAG GCC ACG ACTGC CGA GGA GTT TGA GAT AAT C290
Pfk1Phosphofructokinase, liver, B-typeNM_006828.2GAA GGC CAT GGA CGA GGA GAGGAG CCC CCA CAT TCA GGA TG152
Pgk1Phosphoglycerate kinase 1NM_008828.1GAC TGC ACA CCG AGC CCA TAGCCA CAG TCC AAG CCC ATC CA450
Pkm2Pyruvate kinase, muscle 2NM_011099.2CAT TAC CAG CGA CCC CAC AGCAC GGC ATC CTT ACA CAG CAC226
PtenPhosphatase and tensin homologNM_009775GGC CCA GTC TCT GCA ACC ATGCA GTT AAA TTT GGC GGT GTC A445


The first two authors (M.L.G. and A.V.S.) are acknowledged as equal contributors to this manuscript. This research was funded by NIH grant RO1 AA13205 from the National Institute on Alcohol Abuse and Alcoholism (T.B.K., A.V.S., M.L.G.). K.A.N. was supported on grant T32 ES07282 from the National Institute of Environmental Health Sciences, and K.K.S. acknowledges additional support from grant P30 AA11605. The authors are grateful to expert services at the Microarray Core (University of Louisville, Center for Genetics and Molecular Medicine) and the Bioinformatics, Biostatistics and Computational Biology Core (University of Louisville, Center for Environmental Genomics and Integrative Biology). The authors also thank Jeff Charlap, Reetu Singh and Debbie Dehart for technical assistance. Conflict of Interest Statement: The authors declare they have no competing financial interests.

  • Many abbreviations appear in the literature to denote the 18-kDa mitochondrial peripheral benzodiazepine receptor site, including PBR, MBR, mBzR, pBzR, Bzrp, TspO, and pk18. In the past, we used Bzrp when referring to the PK11195 recognition site specifically (O'Hara et al.,2003). This corresponds to the official gene symbol in mouse (MGI:88222). Recently, new terminology has been introduced renaming it “TranSlocator PrOtein (18 KDa)” (TSPO), which better reflects the current knowledge and scientific evidence regarding the cellular localization and functions of this protein (Papadopoulos et al.,2006). This gene symbol recognizes the protein as a mammalian counterpart to Tryptophan-rich Sensory Protein O, an oxygen-sensing signal generator in protobacteria (Yeliseev et al.,1997).