Participants were right-handed children from the Cape Colored (mixed ancestry) community in Cape Town, South Africa. The Cape Colored community is composed primarily of descendants of white European settlers, Malaysian slaves, Khoi-San aboriginals, and black African ancestors. The incidence of FASD in this population is exceptionally high due to poor socioeconomic circumstances and historical practices of compensating farm labor with wine, which have contributed to a tradition of heavy recreational weekend binge drinking (May et al., 2007, 2013).

Data from 7 of 34 children recruited into the study were excluded from the analyses: technical error (1 control child), incomplete imaging data (2 controls), functional data due to failure to meet the performance criteria (1 FAS, 1 control), and excessive motion (1 FAS, 1 control). Mean and maximum displacements did not differ between exposure groups [t(25) = 1.49, p = 0.162 and t(25) = 1.55, p = 0.148, respectively].

After exclusions, there were 27 children with usable data (11 alcohol exposed and 16 controls). The children with usable scanner data did not differ from those excluded in terms of alcohol exposure or FASD diagnostic group (all ps > 0.20) or other demographic variables, including child age, sex, IQ, and blood lead concentration, and maternal education, age at delivery, parity, and smoking during pregnancy (all ps > 0.10).

13 of the 27 children were the older siblings of participants in our Cape Town Longitudinal Cohort (Jacobson et al., 2008). The others were identified by screening all of the 8- to 12-year-old children from an elementary school in a rural section of Cape Town, where there is a very high incidence of alcohol abuse among local farm workers (Meintjes et al., 2010; Jacobson et al., 2011c).

A research nurse and staff driver transported the mother and child from their home to the Cape Universities Brain Imaging Centre (CUBIC). All examiners were blind with regard to maternal alcohol history and the child’s FASD diagnostic status, except in the most severe cases. Written informed consent was obtained from each mother and assent from each child. Approval for human research was obtained from the ethics committees at Wayne State University and the UCT Faculty of Health Sciences.

Each mother was interviewed in her primary language, Afrikaans or English, regarding her alcohol consumption and smoking during pregnancy. The alcohol interviews used a timeline follow-back approach (Sokol et al., 1985; Jacobson et al., 2002) to determine incidence and amount of drinking on a day-by-day basis during pregnancy. By contrast to most of our Cape Town studies, in which the mothers were interviewed prospectively during pregnancy, the interviews for this sample were conducted retrospectively. Any child whose mother reported consuming at least 14 standard drinks/week (1.0 oz absolute alcohol (AA)/day) on average or engaged in binge drinking during pregnancy (4 or more drinks/occasion) was considered heavily exposed. Controls were children whose mothers reported abstaining or drinking less than seven drinks/week and no binge drinking during pregnancy. Volumes recorded for each type of beverage consumed each day were converted to oz AA using multipliers proposed by Bowman et al. (1975) to provide three continuous measures of drinking during pregnancy: average oz AA/day, AA/drinking day (dose/occasion) and frequency (days/wk). Number of cigarettes smoked on a daily basis was also recorded; use of illicit drugs was obtained as days/month. Mothers were also interviewed regarding their education and occupational status and that of their spouse/partner to determine socioeconomic status (SES) on the Hollingshead (2011) scale. Mothers and children were given breakfast, lunch, and a snack at each laboratory visit. The mother received a small monetary compensation for each visit, as well as a photograph of her child, and the child was given a small gift.

Each child was examined for growth and FAS dysmorphology by two expert United States-based dysmorphologists during a clinic conducted in 2005 (Jacobson et al., 2011c) using a standard protocol based on the Revised Institute of Medicine criteria (Hoyme et al., 2005). The criteria for full FAS were at least two of the principal dysmorphic features, small head circumference (bottom 10th percentile), and low weight or short stature (bottom 10th percentile); for partial FAS (PFAS), two features and small head circumference, low weight, or short stature. Alcohol-exposed children who did not meet criteria for either FAS or PFAS were designated non-syndromal HE.

There was substantial agreement among the examiners on the assessment of all dysmorphic features, including the three-principal fetal alcohol related features—philtrum and vermilion measured on the Astley and Clarren (2001) rating scales and palpebral fissure length (median r = 0.78). Two children who could not attend the 2005 clinic were examined by a Cape Town based FASD expert dysmorphologist (N. Khaole, MD, United States), whose diagnoses were subsequently confirmed by examinations conducted in follow-up clinics we held with the same dysmorphologists in 2009 and with HEH in 2013 and 2016.

Each child was assessed on 7 of the 10 subtests from the Wechsler Intelligence Scale for Children, Third Edition (WISC-III)—Similarities, Arithmetic, Digit Span, Symbol Search, Coding, Block Design, and Picture Completion—and Matrix Reasoning from the WISC-IV. The IQ subtests were selected to represent the four dimensions of the WISC-III: Verbal Comprehension (Similarities), Perceptual Organization (Block Design, Picture Completion, Matrix Reasoning), Freedom from Distractibility (Digit Span, Arithmetic), and Processing Speed (Coding, Symbol Search). Similarities was the only subtest administered in the verbal domain to limit use of tests that may be more dependent on the cross-cultural context. IQ was estimated from these subtests using Sattler’s (1992) formula for computing Short Form IQ; validity coefficients for Short Form IQ based on 5 or more subtests consistently exceed r = 0.90. Handedness was assessed on the Annett (1970a,b) Behavioral Handedness Inventory.

The sample characteristics are summarized in Table 1. The mothers of the alcohol exposed children reported an average of 13.4 standard drinks of alcohol per drinking occasion during pregnancy on 2–3 days per week. By contrast, all but two women in the control group abstained from drinking during pregnancy; the one control mother drank two drinks once during pregnancy and the other control mother drank two drinks about once a month. The groups were generally similar in terms of the other background characteristics, except that mothers of the exposed children smoked more during pregnancy than mothers of controls. None of the mothers in this sample reported using drugs during pregnancy. Of the 11 alcohol-exposed children, 6 met criteria for either FAS or PFAS, while 5 were designated as non-syndromal HE. As expected, the IQ scores of the exposed children were lower than those of the controls. The low IQ scores of both groups reflect the highly disadvantaged circumstances and poor quality of education available in this community.

Behavioral data are summarized in Table 2. Overall accuracy and reaction time did not differ between exposed and control children (all ps > 0.30), nor did accuracy and reaction time at each difficulty level (all ps > 0.25). None of the behavioral measures were related to the continuous measures of prenatal alcohol exposure, although the relation between reaction time in the easy level of the task and AA/occasion fell just short of significance (r = 0.34, p < 0.010), with greater alcohol exposure associated with slower reaction time, as expected.

Performance accuracy was high in both groups, with scores over 80% for all but the most difficult level. A repeated measures ANOVA with Greenhouse–Geisser correction showed that mean accuracy differed significantly between levels [F(1.9,48.7) = 7.60, p = 0.002], with greater accuracy in the easy and intermediate levels than in the difficult level (p = 0.020 and p = 0.005, respectively). Mean reaction times also differed between difficulty levels [F(1.9,49.6) = 9.89, p < 0.0005], with reaction times in the easy condition being faster than in the intermediate (p < 0.013) and difficult (p < 0.001) conditions.

Table 3 shows mean % signal change during non-symbolic number comparison in the Dehaene parietal ROIs for children in each of the two groups. The only significant group difference was in the right PSPL, where the control group showed greater activation than the exposed group, in which activations were actually reduced lower than baseline, on average, however, this effect was no longer significant after adjustment for smoking during pregnancy. By contrast, the left AG showed a stronger distance effect in the exposed children than in the controls (see Table 4).

Since we did not observe a difference in activation during number comparison between exposed and control children in the right IPS as hypothesized, we compared activation levels in this region in the FAS/PFAS group only to the controls. As predicted, the control children activated the right IPS more than the FAS/PFAS group [t(20) = 2.14, p = 0.045, means ± SD = –0.15 ± 0.12 and 0.11 ± 0.28 for the FAS/PFAS and control groups, respectively]. When the FAS/PFAS, non-syndromal HE, and control children were compared, activation patterns in the right PSPL were similar for the FAS/PFAS and HE groups [t(9) = 1.21, p = 0.257], whereas activation patterns in the right IPS of HE children were more similar to those of controls [t(19) = 0.021, p = 0.986].

Higher levels of prenatal alcohol exposure on all three continuous measures were related to greater distance effects in the left AG (rs = 0.46, 0.41, and 0.45 for AA/day, AA/occasion and frequency of drinking, respectively, all ps < 0.05; Figure 3). More drinking days/week was related to reduced activation during number comparison in the right PSPL (r = –0.41, p = 0.036; Figure 4).

Children who showed a greater distance effect in the left PSPL (i.e., activated the left PSPL more with increasing task difficulty) had shorter reaction times (r = –0.43, p = 0.03) and those who showed greater activation of this region had better accuracy (r = 0.46, p = 0.02).

As predicted, prenatal alcohol exposure was associated with weaker activation of the right IPS during non-symbolic number comparison, although the effect was seen only in the syndromal children in this sample. This finding of an adverse effect of prenatal alcohol exposure on activation of the right IPS during non-symbolic number processing is consistent with previous studies of symbolic number processing (Santhanam et al., 2009; Meintjes et al., 2010; Woods et al., 2015). During a subtraction task (Santhanam et al., 2009), adults with alcohol related dysmorphology showed weaker activation of the right inferior parietal lobe (just slightly superior and posterior to our right IPS ROI) than unexposed controls. This effect was not seen in the non-dysmorphic prenatal alcohol exposure group. In a whole brain analysis (Meintjes et al., 2010), we found that control children activated the right IPS more than children with FAS/PFAS during a PJ task, though this region was centered more posteriorly than the right IPS ROI used in this study. There were no non-dysmorphic HE children included in that study. In our previous study of symbolic number processing using the same a priori ROIs as in the present study, we found that increasing alcohol exposure was related to weaker activation in the right IPS during both PJ and simple addition (Woods et al., 2015). These results, taken together with the findings from the current study, indicate that prenatal alcohol exposure impairs the activation of the right IPS during both symbolic and non-symbolic number processing.

The bilateral IPS have frequently been linked to non-verbal representation of quantity (Dehaene et al., 2003). This region is activated when numbers are attended to, even without any explicit number processing task requirements (Eger et al., 2003), during number comparison (Chochon et al., 1999), and during mental arithmetic (Burbaud et al., 1999; Chochon et al., 1999; Pesenti et al., 2000). The region is independent of modality of number and is activated whether number is presented as Arabic numbers, written number words, spoken numbers or sets of dots (Le Clec’H et al., 2000; Pinel et al., 2001; Piazza et al., 2002). It is more active when manipulating large numbers (Kiefer and Dehaene, 1997; Stanescu-Cosson et al., 2000), performing arithmetic with three rather than two operands (Rivera et al., 2002), approximating addition rather than computing the exact sum (Dehaene et al., 1999), and performing subtraction rather than multiplication (Chochon et al., 1999; Lee, 2000). This region has also been shown to exhibit a distance effect, meaning that it is activated more when comparing numbers that are closer together (Dehaene and Cohen, 1996; Pinel et al., 2001). Although we did not observe a behavioral difference in this fMRI study, evidence from our Detroit and Cape Town behavioral studies suggests a specific effect of prenatal alcohol on magnitude comparison (Burden et al., 2005a; Jacobson et al., 2011a,b). The poorer recruitment of the IPS in children with FAS or PFAS observed here provides further evidence of a fetal alcohol exposure related change in mental representation and manipulation of quantity.

The non-dysmorphic HE children did not show poorer recruitment on the IPS than control children. Given that some of the non-dysmorphic children were very heavily exposed, perhaps the effect of prenatal alcohol on the right IPS relates more to the timing of drinking during pregnancy than on the quantity the mother drank.

White matter microstructure directly underlying the left IPS has been shown to be related to mathematical abilities in children with prenatal alcohol exposure. Higher math scores were associated with greater fractional anisotropy in this region (Lebel et al., 2010). These findings, combined with our results and the results of previous fMRI studies, suggest that both gray and white matter of the IPS are involved in mathematical processing and that both are affected by prenatal alcohol exposure.

Based on the findings from developmental dyscalculia (Price et al., 2007; Mussolin et al., 2010a), we expected to find a weaker distance effect in the right IPS in children with prenatal alcohol exposure; however, this was not the case. Although the control children activated the right IPS more than the FAS/PFAS group during non-symbolic number comparison averaged across the three difficulty levels, there was no difference in the distance effect between the groups in this region. The lack of a group difference in the distance effect in this region may be related to the fact that the distance effect was not clearly evident in the right IPS in the control children, possibly because they were too young and their brains still too immature. In a symbolic number comparison task, adults showed a distance effect in parietal regions, including the intraparietal sulcus (Kaufmann et al., 2005), while in an identical paradigm, 8- to 12-year-old children showed no significant distance effect in any parietal region, even though they showed a behavioral distance effect (Kaufmann et al., 2006). An alternative reason may relate to task design. Our “smiley face” stimuli remained the same size throughout trials, so it is possible that children used non-numerical cues, such as density of the faces, or the total area covered by the faces to select the correct answer. Other studies have varied the sizes of their stimuli, so that the participants could not use these non-numerical cues to select the correct answer (Ansari et al., 2006; Holloway and Ansari, 2009; Gullick et al., 2011; Kucian et al., 2011; Lonnemann et al., 2011). A third possibility is that the task did not increase enough in difficulty for the additional intraparietal neuronal resources to be required. However, this is unlikely, as the behavioral results did show distance effects.

Consistent with our hypothesis, exposed children showed compensatory activation of the left AG, demonstrating a greater distance effect in that region than typically developing controls. This finding was also evidenced by an association between level of prenatal alcohol exposure and the distance effect in the left AG. These results suggest that alcohol exposed children need to recruit the left AG to a greater extent as the task difficulty increases, possibly to compensate for deficits in quantity representation in the IPS. In contrast, controls showed an inverse distance effect in this region, that is, reduced left AG activation during the more difficult conditions, presumably due to better functioning of regions specialized for quantity representation. This area has also been implicated in our previous studies of number processing in FASD. In a whole brain voxelwise analysis, we found a significant group difference in a nearly identical region of the left AG [–42,–65,36] with greater activation in the FAS/PFAS group than the controls (Meintjes et al., 2010), and in an ROI study, we found that during PJ children with FAS or PFAS demonstrated greater activation of the left AG ROI than HE or control children (Woods et al., 2015).

The AG is adjacent to the perisylvian language processing network and is associated with the verbal processing of numbers (Dehaene et al., 2003). It is more highly activated during addition and multiplication than during subtraction, presumably because addition and multiplication facts are more likely to be retrieved from long-term memory (Stanescu-Cosson et al., 2000; Simon et al., 2002; Delazer et al., 2005). It is also more active during symbolic number processing than non-symbolic number processing (Holloway et al., 2010; Price and Ansari, 2011). Although the left AG is usually associated with verbal strategies for solving number processing problems, that is not likely to be the case here, as these non-symbolic number comparison problems do not involve the recall of arithmetic facts (Menon et al., 2000), verbal manipulations of number (Dehaene et al., 2003), or the mapping from symbols to numerical magnitudes (Grabner et al., 2007; Ansari, 2008). However, the left AG is also involved in visuospatial attention (Nobre et al., 1997; Corbetta and Shulman, 1998; Gitelman et al., 1999; Corbetta et al., 2000; Rushworth M.F.S. et al., 2001; Rushworth M.F. et al., 2001) and has been shown to be involved in mentally maintaining a spatial representation of numbers similar to a mental number line (Göbel et al., 2001). Based on our data, it appears that the exposed children rely on this spatial representation increasingly as the difficulty of the problems increases, instead of relying on quantity processing mediated by the right IPS.

Exposed children activated the right PSPL less than control children, and more frequent drinking was associated with reduced activation. Although this effect was no longer significant after adjustment for maternal smoking during pregnancy, smoking was confounded with prenatal alcohol exposure in this sample (r = 0.44, p = 0.010), making it difficult to tease out the degree to which each of these exposures was implicated in the observed effect. Prenatal alcohol exposure has not been shown previously to affect the activation of the right PSPL during number processing, but in our previous study we found that greater prenatal alcohol exposure was related to less activation of the left PSPL during EA (Woods et al., 2015). Similarly, a study of subtraction (Santhanam et al., 2009) found that a similar region in the left hemisphere was activated more by controls than by exposed young adults with alcohol related dysmorphology. The PSPL, which is activated during counting (Piazza et al., 2002) and a variety of visual–spatial tasks, is believed to support the engagement of attention during visual processing of numbers (Pinel et al., 2001; Dehaene et al., 2003). These findings suggest that at this age alcohol exposed children seem to be less able to recruit the attentional systems associated with number processing.

Interestingly, activation patterns in the right PSPL for the non-syndromal HE children were similar to those in the FAS/PFAS group, while those in the right IPS were similar to controls. These data suggest that, although the activation of right PSPL appears to be impaired in the non-syndromal HE children, the functioning of the right IPS is apparently spared. Other studies have also found more extensive neural impairment in the FAS/PFAS group than in the HE. For example, both functional and structural connectivity have been found to be lower in heavily alcohol exposed children in only a subset of the regions affected in children with FAS/PFAS (Fan et al., 2016, 2017).

Greater activation of the left PSPL, as well as a greater distance effect on the left PSPL, were both associated with quicker and more accurate behavioral performance. Dehaene et al. (2003) have emphasized that the PSPL can be engaged when attending to specific quantities on the number line. It is possible that the children who are better able to recruit the left PSPL (and recruit it more with increasing task difficulty) are better able to position each array of stimuli on the mental number line and, therefore, to make magnitude comparisons more quickly and accurately.

Because our previous study (Woods et al., 2015) investigating the effect of prenatal alcohol exposure on symbolic number processing in children used the same a priori ROIs as in this study, the results of these two studies can be examined to compare the effect of prenatal alcohol exposure on symbolic and non-symbolic number processing. The tasks investigated in the previous study of 49 children (8–12 years of age) were simple EA and a magnitude comparison task, PJ.

In both studies, although the activation of the right IPS was impaired by prenatal alcohol exposure, we found no relation of alcohol or FASD diagnostic group to activation of the left IPS. It is not clear whether the right IPS is more affected by prenatal alcohol exposure than the left IPS or whether they are equally affected, but only the effect on the right IPS is apparent because the left IPS is recruited to a lesser extent by these tasks. In the Dehaene meta-analysis (Dehaene et al., 2003) used to identify the critical parietal number processing regions used in our studies, the IPS activation was bilateral in all but two studies. To examine whether the absence of an alcohol effect on the left may be due to differences in degree of activation on the left and right, we compared the magnitudes of the activation in the left and right IPS in control children during symbolic and non-symbolic number processing. In control children, neither the left nor the right IPS showed significant activation increases during non-symbolic number processing. In the PJ task, activation increases were seen bilaterally in the IPS (right IPS mean % signal change ± SD = 0.09 ± 0.12, p = 0.007; and left IPS mean % signal change ± SD = 0.07 ± 0.13, p = 0.045), but during EA activation increases were seen only on the right (right IPS mean % signal change ± SD = 0.07 ± 0.08, p = 0.003). It is, thus, not clear whether the left IPS is more sensitive to prenatal alcohol exposure or merely less active in the tasks examined to date.

While during the Smarties and PJ tasks, exposed children showed increased activation of the left AG, no compensatory activation was seen during EA (see also Meintjes et al., 2010). These data suggest that exposed children show compensatory activation of the left AG during magnitude comparison tasks, both symbolic and non-symbolic, but not during arithmetic tasks relying on verbal recall. Whereas for magnitude comparison tasks, the control children appear to use the optimal strategy, relying on the core quantity system in the IPS, the alcohol exposed children recruit the left AG to a greater extent, possibly due to alcohol-related damage to the IPS. In contrast, for arithmetic tasks, verbal recall of number facts (relying on the left AG) is the most efficient strategy and appears to be used by both the exposed and control children.

While prenatal alcohol exposure was associated with lower activation of the right PSPL during the Smarties task, no association with alcohol exposure was found in this region during either PJ or EA (see also Santhanam et al., 2009; Meintjes et al., 2010). In contrast, alcohol-related reductions were seen in activation in the left PSPL during EA (Woods et al., 2015) and in a comparable region during subtraction (Santhanam et al., 2009). Thus, it appears that both the left and right PSPL are impaired by prenatal alcohol exposure, but the impairment of the right PSPL is evident only during non-symbolic magnitude comparison, whereas impairment of the left PSPL has been seen during symbolic arithmetic operations. Neither of the studies using the PJ task to assess symbolic magnitude comparison found alcohol-related impairment in the left nor the right PSPL.

Although all but one of the heavily alcohol exposed children showed relatively low levels of activation in the right PSPL during the Smarties task, there was a notably wide range of signal change among the controls (see Figure 4). We examined numerous sample characteristics (child IQ, age at scan, grade in school, ADHD status, maternal education and SES, smoking and other prenatal drug exposures, and postnatal lead exposure) in relation to these activation levels in the control group. One of the two controls with the lowest % signal change had a somewhat elevated prenatal exposure to smoking (seven cigarettes/day). There were no other differences that would help explain the wide range in responses, suggesting that these activations represent a “normal” range for the controls. What is striking in our findings is that the activation levels in this brain region are low for virtually all of the alcohol-exposed children.

One limitation of this study was that, by contrast to most of the studies our group has conducted in Cape Town, the maternal report of drinking during pregnancy for this cohort was obtained retrospectively several years after the child’s birth. Nevertheless, the validity of these reports is supported by the fact that they were predictive of neuroimaging and neurobehavioral outcomes (Meintjes et al., 2010, 2014; Jacobson et al., 2011b,c; De Guio et al., 2014; du Plessis et al., 2014; Lewis et al., 2015; Lindinger et al., 2016). Predictive validity of the maternal pregnancy drinking reports for childhood IQ in this sample was substantial (rs = –0.54 and –0.53, for AA/day and AA/occasion, respectively, both ps < 0.001). However, drinking during pregnancy is more difficult to recall reliably in a retrospective interview than smoking, which is a more addictive, more consistent, and less of an episodic practice, as well as packaged so that quantity is more readily recalled (Jacobson and Jacobson, 1992). These differences in ability to recall quantity between alcohol exposure and smoking make it particularly difficult to tease out the degree to which each of these exposures is responsible for the outcomes, such as activation of the right PSPL, that are related to both exposures in this sample. A second limitation was the small size of the FAS/PFAS (n = 6) and HE (n = 5) groups. In addition, because all of the children come from socioeconomically and educationally disadvantaged environments, we cannot determine the degree to which the results would hold for children from an educationally less deprived background. We did not control for multiple comparisons, due to the fact that we examined only five regions rather than the whole brain; ROI analyses increase SNR by averaging across the voxels in a region. Although the use of an ROI approach was appropriate for this study, a whole brain voxelwise analysis could have revealed additional activation differences between groups.