Prenatal alcohol exposure (PAE) is associated with a range of neurocognitive and behavioral problems. Fetal alcohol spectrum disorders (FASD) is an umbrella term under which the entire spectrum of outcomes related to PAE falls. Fetal alcohol syndrome (FAS), the most severe of the FASD, is characterized by growth deficits, small head circumference, distinctive craniofacial dysmorphology (small palpebral fissures, flattened philtrum, and thin vermillion) and cognitive problems that relate to specific brain abnormalities (Jones and Smith, 1973; Hoyme et al., 2005). Partial FAS (PFAS) is diagnosed in individuals exhibiting the facial dysmorphology characteristic of FAS, whose mothers are known to have drunk heavily during pregnancy and who exhibit growth deficits, small head circumference or neurobehavioral impairment (Hoyme et al., 2005). Adverse effects on brain development range from microstructural, neurochemical and cellular dysfunction (such as widespread apoptotic neurodegeneration that can result in the death of millions of cells in the forebrain during gestation) to gross structural abnormalities (Ikonomidou, 2000; Olney, 2004; Spadoni et al., 2007; Lebel et al., 2008).

In the U.S. and in Western countries, FAS occurs at a rate estimated to be between 0.05 and 7 per 1,000 live births (May and Gossage, 2001; May et al., 2009). In contrast, certain high-risk communities in South Africa have been found to have an FAS prevalence rating that is 18–148 times greater than in the U.S. (May et al., 2000; Viljoen et al., 2005; Urban et al., 2008). Within the Western Cape province of South Africa, rates are highest amongst the Colored population of mixed Asian, African and European ancestry (May et al., 2000, 2005, 2013; Viljoen et al., 2005).

Magnetic resonance imaging (MRI) allows for non-invasive and quantitative investigation of the structural and functional brain alterations that are observed in FASD. Brain regions that have been consistently implicated in FASD include the corpus callosum (CC), cerebellum, parietal lobe, frontal lobe, temporal lobe, caudate nucleus, hippocampus and increased volume of lateral ventricles (see review papers, Lebel et al., 2011; Donald et al., 2015). While most regions show alcohol-related volume reductions, findings in certain subcortical brain regions have not always been consistent. For example, some studies report bilateral decreases in hippocampal volume (Willoughby et al., 2008; Astley et al., 2009; Coles et al., 2011; Nardelli et al., 2011; Treit et al., 2013) with PAE, while others find relative sparing (Archibald et al., 2001), or even volume increases (Riikonen et al., 2007). Methodological differences, such as automated vs. manual segmentation, failure or different strategies to control for total intracranial volume (TIV), small sample sizes, or large age ranges may explain some of these differences. Notably, one study found increased hippocampal volume relative to brain volume with age (9–15 years) in controls, but not in children with FASD (Willoughby et al., 2008), highlighting the potential contribution of age-related changes in brain development when comparing diagnostic groups and the need for narrow age ranges. It is not yet known to what extent developmental trajectories in children prenatally exposed to alcohol are altered.

Although manual tracing is considered the gold standard for volumetric analysis, yielding the greatest consistency and sensitivity (Morey et al., 2009), few studies employ this method due to high time demands (Archibald et al., 2001; Riikonen et al., 2007; Willoughby et al., 2008; Astley et al., 2009). Here we use manual tracing of study participants within a narrow age range (10.4 ± 0.4 years) to examine the effects of PAE on volumes of select subcortical structures and the CC in 81 children from the Cape Town Longitudinal Cohort (Jacobson et al., 2008), for whom diagnostic and detailed prospective prenatal alcohol and drug exposure data are available. Most previous studies have only been able to examine volumetric differences between diagnostic groups and not dose-dependent effects, since the amount of alcohol used during pregnancy is difficult to recall reliably in retrospective case-controlled studies (Jacobson et al., 2002).

Due to the labor intensive nature of manual tracing, only subcortical structures previously implicated in FASD were investigated (Archibald et al., 2001; Morey et al., 2009; Nardelli et al., 2011). These include the caudate nucleus and hippocampus, which are involved in cognition, memory and emotional networks (Devinsky and D'Esposito, 2004) that are amongst the most affected domains in FASD (Uecker and Nadel, 1996; Mattson et al., 1998; Willoughby et al., 2008; Green et al., 2009; Jacobson et al., 2011; Lewis et al., 2016). Although findings in the nucleus accumbens have rarely been reported in FASD (Archibald et al., 2001), we report it here because the manual tracing protocol we used for the caudate initially involved tracing both structures together. The CC was included as it is clearly visible on MRIs, is easy to delineate, and has been consistently shown to be reduced in FASD (Mattson et al., 1992; Sowell et al., 2001, 2008; Bookstein et al., 2002; Autti-Rämö et al., 2007; Lebel et al., 2008; Astley et al., 2009), even in newborns (Jacobson et al., 2017). PAE has been linked to a larger angle of the splenium (Bookstein et al., 2007), reductions in CC thickness and area in the splenium and anterior third (Yang et al., 2012), and CC volume reductions in the splenium, genu and area just anterior to the splenium (Riley et al., 1995).

In addition, we examined whether alcohol-related regional volumetric changes mediate the effect of PAE on cognitive performance on the Wechsler Intelligence Scale for Children-Fourth Edition (WISC-IV) and the California Verbal Learning Test-Children's Version (CVLT-C). Poorer IQ is frequently seen in FASD (Streissguth et al., 1990, 1991; Jacobson et al., 2004) as is impaired performance on the CVLT-C (Delis et al., 1994), a list learning task (Kerns et al., 1997; Mattson et al., 1998; Crocker et al., 2011; Vaurio et al., 2011; Suttie et al., 2013; Lewis et al., 2015).

We hypothesized that increasing PAE would be associated with volumetric reductions in all regions examined and that these effects would remain significant after controlling for reductions in total brain volume. We also hypothesized that volumetric reductions in hippocampus, caudate nucleus and CC would be associated with poorer WISC IQ and CVLT-C performance and that the effects of PAE on these outcomes would be partially mediated by volumetric changes in the regions that were traced.

Following the exclusions described above, we report data for 71 children. Demographic characteristics are presented in Table 2. Although the FASD diagnostic groups did not differ by sex, prenatal exposure to cigarettes, marijuana, or cocaine, or childhood lead exposure, children in the HE group were slightly older than those in the FAS and PFAS groups. As expected, children in both the FAS and PFAS groups had lower IQs than the HE children; those in the PFAS group also had lower scores than the controls. Total brain volumes were smaller in the children with FAS compared to all other groups. SES was lower for mothers of children with FAS or PFAS than mothers of children from the HE and control groups.

As expected, all of the alcohol measures for the control group were statistically lower compared to the alcohol-using groups (Table 2). In addition, children with FAS were more exposed than HE children on all measures. They were also exposed to more alcohol/drinking day at conception than those in the PFAS group. At conception and across pregnancy, AA/day and drinking days/week for the PFAS group were also significantly higher than for the HE group. Although number of days of alcohol use was reduced from 1.5 (at conception) to 1.1 (across pregnancy) day/week on average [t₍₆₈₎ = 3.88, p < 0.001), the number of drinks/occasion remained virtually unchanged, [t₍₆₉₎ = −0.33, p = 0.741]. Moreover, alcohol consumption at conception was highly correlated with alcohol use throughout pregnancy (r = 0.923). Due to the small sample size of the FAS group, the two syndromal groups, FAS and PFAS, were combined for subsequent volumetric analyses.

Volumes of traced regions are reported for each diagnostic group in Table 3 and associations of control variables with structural volumes in Table 4. Children in the FAS/PFAS group had smaller right hippocampi compared to both the HE and control children, but this effect was no longer significant after adjustment for TIV. Volumes of four regions were smaller in girls than in boys. Among the remaining potential confounders, smoking during pregnancy and postnatal lead exposure were related to size in one region—the CC.

Table 5 presents the results of the multiple regression analyses examining the relation of continuous measures of PAE to the structural volumes. PAE was associated with volume reductions bilaterally in the caudate and hippocampus and smaller CC, but the effect on the CC was no longer significant after controlling for smoking during pregnancy and postnatal lead exposure. When the regression for CC was rerun controlling only for lead, the effect of absolute alcohol/day remained significant (β = −0.281, p = 0.022). In a regression of the effect of alcohol and smoking on CC volume, the effects of both of these exposures fell just short of significance (alcohol β = −0.236, p = 0.065; smoking β = −0.217, p = 0.089), suggesting that neither effect was attributable to the effect of the other exposure.

Prenatal alcohol exposure (PAE) was associated with lower IQ (r = −0.402; p = 0.001) and CVLT-C (r = −0.248; p = 0.045) scores. Larger CC was associated with higher IQ for the sample as a whole and with better CVLT-C performance for the FAS/PFAS group (Table 6). Unexpectedly, smaller hippocampi were associated with better CVLT-C scores for the FAS/PFAS group.

The hypothesis that CC size would mediate the effect of PAE on IQ was tested using multiple regression. PAE was entered in Step 1. When CC size was added in Step 2 of the regression analysis, the effect of AA/day on IQ was reduced from −0.40 to −0.33, a reduction that was statistically significant, Clogg (t = −2.86, p < 0.01). These data thus indicate that the effect of PAE on IQ was partially mediated by the alcohol-related reduction in CC size (Figure 2).

This study shows that increasing PAE, both in terms of quantity and frequency, is associated with disproportionate volume reductions bilaterally in the caudate nuclei and hippocampi and smaller CC. By contrast, volume differences between FASD diagnostic groups were evident only in the right hippocampus. Smaller CC was associated with poorer IQ. Within the FAS/PFAS group, smaller CC was also associated with lower CVLT-C scores, but hippocampal volume reductions were associated with better CVLT-C performance. The results from our mediation analysis suggest that the adverse effect of PAE on IQ is partially attributable to the reductions in CC size.

When the regional volumetric measures were compared across the three diagnostic groups, no significant differences were seen after adjustment for TIV. By contrast, the continuous measures of PAE were associated with volumetric differences in all of the regions examined after adjustment for brain size, except the bilateral NA. These findings are consistent with other neuroimaging studies performed with children from the same Cape Town cohort, which found that continuous measures of PAE were more sensitive than diagnosis to fetal alcohol-related alterations in brain structure and function (De Guio et al., 2014; Meintjes et al., 2014; du Plessis et al., 2015).

Our finding of alcohol-related volumetric reductions bilaterally in the caudate nuclei are consistent with studies reporting decreased basal ganglia volume in FAS (Mattson et al., 1994, 1996, 1999, 2001; Archibald et al., 2001; Roussotte et al., 2012). Animal studies suggest that the physiological mechanisms underlying smaller striatal volumes may be explained by a reduction in dendritic tree formation rather than a reduction in number of parvalbumin interneurons (De Giorgio et al., 2012). The caudate nucleus is involved in many of the neurobehavioural deficits exhibited by children prenatally exposed to alcohol, including executive function domains, such as cognitive flexibility, concept formation and reasoning, planning and response inhibition (Mattson et al., 1999, 2001). This region has also been shown to be activated less by children with FASD during behavioral inhibition tasks (Fryer et al., 2007) and to mediate cognitive control and verbal learning in children with PAE (Fryer et al., 2012). Although the latter study reported that larger caudate nuclei were associated with better performance in verbal learning in alcohol-exposed children, we did not confirm that finding in our study.

The caudate nuclei may be more susceptible to the effects of PAE than the NA. It has been suggested that some midline brain structures are more affected by alcohol exposure during embryological development than others (Sulik et al., 1981; Zhou et al., 2003; Meintjes et al., 2014). Our results are consistent with those of Archibald et al. (2001), who also failed to find alcohol-related volume reductions in the NA using a similar tracing methodology. Because the NA can only truly be differentiated from the caudate nucleus histologically, the lack of association in both these studies may be due to poor sensitivity by these tracing methods to differentiate the NA from the caudate nuclei. Thus, although our results suggest that the NA may be relatively spared in PAE, further investigation using more sensitive methods, such as diffusion tensor imaging or higher resolution structural imaging, is needed.

Prenatal alcohol exposure (PAE) was also associated with volumetric reductions in both the left and right hippocampi–effects that remained significant after control for TIV. Reductions in pyramidal and granule cell number and density were observed in rats exposed prenatally to alcohol (Livy et al., 2003), which suggests that this may be the case in humans. The hippocampal formation has been shown to be particularly sensitive to PAE, showing a diverse range of negative effects and is one of the few areas of the brain which continues to exhibit these deficits as it develops into adulthood (see review, Gil-Mohapel et al., 2010). Associations of smaller hippocampi with better memory performance in typically-developing young adults (Van Petten et al., 2004) suggest a neural pruning process with age that may alter the relation of hippocampal volume in healthy controls as they grow older. Although our findings for the caudate nuclei, NA and CC are consistent with those of Archibald et al. (2001), who used a similar tracing protocol to ours, fetal alcohol-related differences in hippocampal volumes in their study did not survive after controlling for TIV. This disparity in results may be due to the wide range of ages included in that study (7–24 years) compared to the more limited range in ours (9–11 years). The observed association of smaller hippocampal volume with better learning and memory performance in the FAS/PFAS group in the current study was not expected at this age and is difficult to interpret.

Prenatal alcohol exposure (PAE) was associated with smaller CC after adjustment for TIV and postnatal lead exposure. However, regression analyses showed that the effects of smoking and alcohol on CC volume were confounded in this sample, and both fell slightly short of conventional levels of statistical significance when included in the same analysis. The effects of maternal smoking on the child's brain are not well understood, but decreases in CC size and changes in microstructure of other frontal WM structures have been linked with maternal smoking during pregnancy (Jacobsen et al., 2007; Jacobson et al., 2008; Paus et al., 2008). Given that the PAE and maternal smoking effects were similar in magnitude when adjusted for each other in the regression analysis, neither effect appears to be attributable to the effect of the other exposure, and a larger sample would likely have shown their effects to be independent.

The hypothesis that volumetric reductions in the caudate nuclei and hippocampi would mediate effects of PAE on cognitive performance was not supported. Volumetric reductions in the caudate nuclei did not relate to the cognitive outcomes; and, contrary to expectation, the hippocampal volumes in the FAS/PFAS group were inversely related to learning and memory performance.

In this study, larger CC area was associated with higher IQ scores and, amongst children in the FAS/PFAS group, with better learning and memory performance. Mediation analysis indicated that the effect of PAE on IQ was partially mediated by reductions in CC size. Our findings are consistent with several studies that have suggested that WM is more susceptible to the teratogenic effects of PAE than cortical GM (Archibald et al., 2001; Lebel et al., 2008; Spottiswoode et al., 2011). It is of interest that microstructural changes in two regions of the CC have also been found to mediate effects of PAE on IQ (Fan et al., 2016).

Morphology of the CC has been shown to have functional effects, with a thinner midline being associated with poorer motor activity and thicker than average midline being associated with poorer executive function (Bookstein et al., 2002). The CC is particularly important for functional outcomes requiring interhemispheric transfer of information, and it has been shown that even moderate exposure to alcohol may influence this process (Roebuck et al., 2002; Dodge et al., 2009). Since cognitive function assessed on IQ tests depends on a broad range of cognitive domains (Wechsler, 2003; Willoughby et al., 2008) utilizing a range of GM areas that communicate via axons in the CC, poorer interhemispheric transfer could contribute to lower IQ (Dodge et al., 2009). Smaller CC size may be due to either poorer myelination or fewer axons in the CC, resulting in slower signal processing or communication with greater numbers of GM cell bodies, respectively. CC volume reductions may also be due to changes to CC projection neurons, as reductions in size, number and location of these cells were observed in rats prenatally exposed to alcohol (Livy and Elberger, 2008). However, as no one animal model exhibits all the features of FASD (Cudd, 2005), further investigation into the underlying physiological mechanisms in humans is needed to better understand the links between CC and behavioral outcomes. As such, it is not surprising that the most heavily exposed children, those in the FAS/PFAS group, had the smallest CC areas and the lowest IQ scores.

One potential limitation of this study that is common to other longitudinal PAE studies is that it relies on the mother's report to assess alcohol consumption. However, we have previously validated the TLFB interview used in this study in relation to levels of fatty acid ethyl ester metabolites in meconium samples obtained from newborns in this cohort (Bearer et al., 2003). In addition, this TLFB interview has been shown to predict a broad range of behavioral (Jacobson et al., 2002, 2008; Lewis et al., 2015; Lindinger et al., 2016; Fortin et al., 2017) and neuroimaging outcomes (De Guio et al., 2014; Meintjes et al., 2014; du Plessis et al., 2015; Jacobson et al., 2017). A second limitation relates to the difficulty assessing effects of timing of exposure in a human study. Animal models have demonstrated that timing of exposure is a major determinant of specific longterm neurobehavioral outcomes, since different brain regions are susceptible to PAE during unique sensitive periods in their development (Livy et al., 2003; Valenzuela et al., 2012; Sadrian et al., 2014). Unfortunately, multicollinearity between maternal alcohol use at conception and later in pregnancy in this cohort, as in most human samples, was too high to discriminate effects attributable to timing of exposure. Polysubstance abuse is another common limitation in PAE studies. The current study assessed potential confounding by prenatal cocaine, marijuana and smoking exposure in order to address this problem; none of these exposures were related to the outcomes examined. Strengths of this study include the use of a sample size that was large relative to other manual tracing studies, limited age range, a population that is geographically and culturally homogenous albeit diverse in terms of genetic ancestry, and the availability of continuous measures of prospectively collected alcohol consumption.

In conclusion, this study, which used manual tracing, confirms previous reports linking PAE to decreases in subcortical volumes in the caudate nucleus and hippocampus after adjustment for TIV and also showed that PAE is associated with a disproportionate reduction in CC size. This is the first study to provide evidence that fetal alcohol-related reductions in CC partially mediate effects of PAE on IQ. This finding adds to the growing body of evidence suggesting that WM may be particularly vulnerable to the teratogenic effects of PAE on development of the fetal brain.

Ethical approval for the study was acquired from the University of Cape Town Faculty of Health Sciences Human Research Ethics Committee and the Wayne State University Institutional Review Board. Informed written consent was obtained from the mothers and oral assent from the children in accordance with the Declaration of Helsinki.

SB performed the tracing of the volumes under the supervision of CW and with the assistance of SR. She also reviewed the literature, performed the data analyses, contributed to the interpretation and wrote up the findings. EM provided overall project supervision, collaborated on the design of the neuroimaging study, the data analysis, interpretation and write-up of the findings. SJ and JJ designed the study, supervised recruitment and maternal interviews and child assessments, collaborated on the data analysis, the interpretation of the findings and write-up of the paper. CM administered the maternal interviews, which included sociodemographic information and alcohol, smoking and drug ascertainment. NL and CL performed neuropsychological assessments.