Birth Weight, Head Circumference, and Prenatal Exposure to Acrylamide from Maternal Diet: The European Prospective Mother–Child Study (NewGeneris)

Study population. The study was conducted by the NewGeneris consortium (www.newgeneris.org) as part of research exploring the impact of diet during pregnancy on child health (Merlo et al. 2009). Pregnant women enrolled during 2006–2010 in 11 maternity units located in Copenhagen, Denmark; Heraklion, Greece; Oslo and Akershus, Norway; Barcelona and Sabadell, Spain; and Bradford, England (Chatzi et al. 2012; Magnus et al. 2006; Pedersen et al. 2009; Raynor et al. 2008). Specific eligibility criteria were applied in the baseline cohorts for the participation of mothers [for details, see Supplemental Material, Table S1 (https://doi.org/10.1289/ehp.1205327)]. Precise participation rates for the present analysis cannot be estimated because several filters were applied for inclusion of mothers in the NewGeneris study, such as giving birth during the periods of cord blood collection and processing, getting sufficient volume of cord blood, successful blood processing, and analyzing biomarkers.

Questionnaire information on maternal characteristics and cord blood Hb adduct measurements was available for 1,151 mother–child pairs. We excluded 16 pairs of twins and 34 mother–child pairs with missing information on maternal smoking, gestational age, birth weight, and/or sex of the child, leaving 1,101 mother–child pairs for analysis. Information on birth weight, head circumference, gestational age, sex, and mode of delivery was obtained from maternity records.

Gestational age for participants from Denmark, Greece, and Spain was estimated from the interval between last menstrual period and date of the delivery and was corrected by ultrasound measurement if there was a discordance of ≥ 7 days between both estimates. Ultrasound-based estimation was provided for most participants from England and Norway. We defined small-for-gestational-age (SGA) children as those who weighed less than the 10th percentile of the cohort-specific reference of fetal growth (mean birth weight for all SGA children: 2,838 g), stratified by completed week of gestation and sex.

Ethical approval was obtained from the research ethics committee in each country: the Regional Ethical Review Board in Stockholm, Sweden; the Capital Region of Denmark; the Ethical Committee of the University Hospital in Heraklion, Crete, Greece; the Norwegian Regional Committee for Medical and Health Research Ethics; the Clinical Research Ethics Committee of Barcelona, Spain; and the Bradford Local Research Ethics Committee, Bradford, United Kingdom. Further details regarding ethics approval are provided in Supplemental Material (https://doi.org/10.1289/ehp.1205327). Written informed consent for participation of the women and their children was obtained from all participating women.

Acrylamide and glycidamide Hb adducts in cord and maternal blood. Cord blood was collected in heparin tubes from the placenta by umbilical puncture immediately after delivery, following a common protocol at each site. Erythrocytes were separated by centrifugation and stored at –20°C before and after shipment on dry ice. Hb adducts from acrylamide and glycidamide were simultaneously determined by the “adduct FIRE procedure” (fluorescein isothiocyanate R Edman) in 1,101 cord blood samples and 172 maternal blood samples (von Stedingk et al. 2010, 2011). In brief, adducted N-terminal valines were detached using the Edman reagent fluoresceine-5-isothiocyanate. The detached analytes were purified by solid phase extraction and analyzed by liquid chromatography/mass spectrometry.

Ethylene oxide Hb adduct levels that are associated with exposure to tobacco smoke were also measured in 1,074 cord blood samples (to assess exposure to tobacco smoke during pregnancy) using the same methods (von Stedingk et al. 2011).

Maternal diet. Detailed information on maternal diet during pregnancy was obtained from the mothers using FFQs collected before or at the time of delivery (Merlo et al. 2009). We grouped food and drink items into eight groups known to contain potentially high levels of acrylamide based on similarities in the composition and processing of the foods in each group—specifically, fried potatoes, potato chips, breakfast cereals, crisp bread, coffee, cookies, fine bakery products, bread and toast (Brantsaeter et al. 2008; EFSA 2010; JECFA 2011). We developed an acrylamide food score for nonsmokers following the same approach used for the evaluation of the Mediterranean diet (Trichopoulou et al. 2003). The score was based on consumption of the eight food groups (as mentioned above); each nonsmoking woman received a score of 0 for each food if her consumption was below the country-specific median, and a score of 1 if her consumption was above the median. Scores were then added for each woman and ranged from 0 (lowest intake) to 8 (highest). In all dietary analyses we excluded women with missing FFQs (n = 146), those who smoked during the last 4 months of pregnancy (n = 129), and 25 women with a total energy estimate of < 500 or > 6,000 kcal/day (Butte and King 2005; Willett 1998).

Statistical analysis. We used linear regression models for birth outcomes evaluated on a continuous scale (birth weight, head circumference) and calculated beta-coefficients and 95% confidence intervals (CIs). Logarithmic-binomial regression models were used to determine relative risks (RRs) and 95% CIs for SGA. Cord blood Hb adduct levels were modeled on a continuous scale (effect estimated per 10 pmol/g Hb increments) or categorized according to quartiles of the distribution. All models were adjusted for country of the child’s birth and gestational age (continuous in completed weeks). Additional adjustment for gestational age as a quadratic term (continuous completed weeks squared) gave nearly identical results (not shown). Further adjustment was based on a priori selection of potential risk factors for reduced birth weight including maternal smoking (no, yes), passive smoking (no, yes), sex, prepregnancy body mass index (BMI; kilograms per meter squared), parity (0, ≥ 1), maternal age (years), maternal ethnicity (white, nonwhite), and maternal education (low, middle, high) as a marker of socioeconomic position. In addition, models were adjusted for high or low maternal consumption during pregnancy (according to country-specific median intake) of fruit and vegetables (based on 20–61 questionnaire items), fish (fresh, canned, dried, and prepared fish and shellfish, including items with high and low fat content), and soft drinks (e.g., cola, other soft drinks, energy soft drinks including regular and light products) as markers of more versus less healthy dietary patterns.

The average lifespan of erythrocytes is 4 months in adults (Törnqvist et al. 2002). Therefore, women who never smoked or who quit smoking before the last 4 months of pregnancy were categorized as “nonsmokers,” and those who never smoked were categorized as “never smokers.” All associations were examined for the full study population (n = 1,101) and separately for children born to nonsmokers (n = 972).

In addition to estimating associations adjusted for country, we estimated country-specific associations between acrylamide Hb adducts and birth weight, and performed a meta-analysis to derive pooled estimates of effect. Only results from fixed-effects models are shown because heterogeneity of effects between centers were not statistically significant.

Associations of maternal diet with acrylamide Hb adduct levels in cord blood, and with birth weight, were estimated using linear regression models that were adjusted as described above. Exposure–response curves were estimated using a generalized additive model with the dietary score modeled as a smoothed spline with 2 degrees of freedom, adjusted for country and gestational age. We used alpha levels of 5% as reference value for statistical significance and Stata S.E. version 10.0 for the statistical analyses (StataCorp, College Station, TX, USA).

Levels of Hb adducts from acrylamide. Acrylamide and glycidamide Hb adducts in cord blood were detectable in all children (n = 1,101). The median Hb adduct level from acrylamide was 14.4 pmol/g Hb (range, 4.4–147.6) and for glycidamide was 10.8 pmol/g Hb (range, 2.0–117.6). There was a statistically significant correlation between glycidamide and acrylamide adduct levels in cord blood (Pearson correlation coefficient r = 0.85, p < 0.001). Median acrylamide adduct levels were higher in cord blood from children of mothers who smoked (n = 129) than in children of nonsmokers (n = 972; 30.5 vs. 13.8 pmol/g Hb, p < 0.001). Corresponding levels for glycidamide adducts were 20.7 versus 10.1 pmol/g Hb (p < 0.001), and for ethylene oxide adducts (as a marker of exposure to cigarette smoke) were 24.5 versus 8.9 pmol/g Hb, (p < 0.001).

The median acrylamide Hb adduct levels in cord blood were approximately half of the levels in paired maternal blood (Figure 1). Hb adduct levels in cord blood were positively correlated with both maternal acrylamide (r = 0.95, p < 0.001, n = 171) and glycidamide Hb adducts (r = 0.94, p < 0.001, n = 171). Among nonsmokers the highest median level of acrylamide adducts was detected in children from England (23.6 pmol/g Hb) and the lowest in children from Denmark (12.0 pmol/g Hb).

Factors associated with birth weight. The participating mothers were mainly white, nonsmoking, in their late 20s to early 30s with prepregnancy BMI in a range of 18.5–24.9 kg/m² and already parous (Table 1). The prevalence of smokers was lowest in Norway and Denmark (2.4% and 3.8%, respectively) and highest in Greece (21%). Birth weight was significantly increased in association with maternal age, parity (parous vs. nulliparous), prepregnancy BMI, nonsmoking status, male sex, and gestational age (Table 2). On average, children from the Northern European countries weighed more than children from Spain and Greece (p < 0.001).

Hb adduct levels and birth weight. Higher levels of acrylamide and glycidamide adducts in cord blood were associated with a significant decrease in birth weight (Table 3). The mean birth weight was reduced by 35 g (95% CI: –51, –19 g) with each 10-pmol/g Hb increase in acrylamide adduct levels after adjusting for country of birth and gestational age in the total population, and by 20 g (95% CI: –46, 6 g) among 972 nonsmokers. A 10-pmol/g Hb increase in glycidamide adducts was associated with a 60-g reduction in birth weight (95% CI: –87, –34 g) in the total population, and a 53-g reduction (95% CI: –95, –10 g) in nonsmokers. Results were similar among the subgroup of 889 women who were never-smokers, with a mean birth weight reduction of 26 g (95% CI: –54, 3 g; p = 0.074) for a 10-pmol/g Hb increase in acrylamide adducts when adjusted for gestational age and country of birth; the corresponding reduction in birth weight for glycidamide adducts was 65 g (95% CI: –111, –19 g).

Birth weight decreased monotonically with increasing quartiles of exposure (Table 3). The estimated difference in birth weight for infants in the highest versus lowest quartile of acrylamide Hb adduct levels after adjusting for gestational age and country was –132 g (95% CI: –207, –56 g) in the total population and –107 g (95% CI: –188, –27 g) when restricted to nonsmokers. The estimated difference in birth weight for infants in the highest versus lowest quartile of glycidamide adducts was –136 g (95% CI: –212, –60 g) in the total population and –103 g (95% CI: –182, –23 g) among nonsmokers (Table 3). Among term deliveries (n = 1,063), the difference in birth weight for children in the highest versus lowest quartile of acrylamide Hb adduct levels was –137 g (95% CI: –214, –60 g); a similar difference was observed for 939 term babies of nonsmokers (–120 g; 95% CI: –201, –38 g).

Associations between Hb adducts and birth weight adjusted for country and gestational age were not substantially modified when further adjustments were made for other potential risk factors for reduced birth weight, including sex, prepregnancy BMI, parity, maternal age, ethnicity, education, passive smoking and smoking, and dietary variables that could be related with a healthy (or unhealthy) eating pattern such as intakes of vegetables and fruits, fish, or soft drinks (Table 3). The population in the comprehensively adjusted models (n = 747) is smaller than the population in models adjusted only for country and gestational age (n = 1,101). We also estimated associations adjusted for ethylene oxide Hb adduct levels in cord blood as a biomarker for active/passive smoking, in addition to the other covariates listed above. The difference in birth weight for infants of nonsmokers (n = 656) in the highest versus lowest quartile of acrylamide Hb adduct levels remained statistically significant after adjustment for ethylene oxide adducts (–142 g; 95% CI: –246, –38 g), whereas the corresponding difference was borderline significant for glycidamide Hb adducts (–82 g; 95% CI: –182, 17 g; p = 0.11).

Negative associations between acrylamide and glycidamide Hb adduct levels in cord blood and birth weight were observed for all countries (Figure 2). A meta-analysis of country-specific estimates gave similar results to estimates based on combined data adjusted for country, with a 10-pmol/g Hb increase in acrylamide adducts associated with a 36-g (95% CI: –53, –19 g) decrease in birth weight, and a 10-pmol/g Hb increase in glycidamide adduct levels associated with a 63-g (95% CI: –90, –36 g) decrease in birth weight.

The pattern was also consistent among nonsmokers (–26 g, 95% CI: –52, 1 g, and –63 g, 95% CI: –106, –20 g for a 10-pmol/g Hb increase in acrylamide and glycidamide, respectively), with no significant heterogeneity between countries (p = 0.51 and p = 0.46), although in England the association with birth weight was minimal (–2 g, 95% CI: –44, 41 g, and –18 g, 95% CI: –95, 59 g).

Hb adduct levels and SGA. SGA (birth weight < 10th percentile for the cohort according to week of gestation and sex) was increased in association with a 10-pmol/g Hb increase in acrylamide adduct levels in the full population (RR = 1.20; 95% CI: 1.08, 1.33, based on 891 observations and 72 SGA births) and for infants of nonsmokers (RR = 1.35; 95% CI: 1.10, 1.65, based on 794 observations and 60 SGA births). The corresponding estimates for glycidamide were RR = 1.36 (95% CI: 1.13,1.64) for all and RR = 1.42 (95% CI: 1.00, 2.02) for nonsmokers.

Hb adduct levels and birth head circumference. The highest versus lowest quartile of acrylamide Hb adduct levels was associated with a significant reduction in head circumference of 0.33 cm (95% CI: –0.61, –0.06 cm) in the full population and among nonsmokers, with similar results observed for glycidamide (Table 4). Similar to the associations with birth weight, there was a monotonic reduction in birth head circumference with increasing quartiles of exposure. This pattern was also found after further adjustment in the smaller study population with available information on potential risk factors (Table 4). These associations, however, were not statistically significant when adjusted for the other potential risk factors, as above.

Maternal diet, acrylamide Hb adduct levels in cord blood, and birth weight. A 1-unit increase in the acrylamide-rich food score was associated with higher Hb cord blood adduct levels (Figure 3) for acrylamide (0.68 pmol/g Hb; 95% CI: 0.30, 1.06) and glycidamide (0.39 pmol/g Hb; 95% CI: 0.15, 0.63) based on food score modeled as a simple continuous variable. Consistent with the associations observed between Hb adducts from acrylamide and birth weight, a 1-unit increase in the acrylamide food score was associated with a 16-g decrease in birth weight (95% CI: –33, 1; p = 0.066) after adjustment for country and gestational age. Additional adjustment did not change this association (Figure 3). Higher food scores were associated with a nonsignificant reduction in birth head circumference of –0.01 cm (95% CI: –0.07, 0.05 cm; p = 0.72) among children of nonsmoking women with acrylamide food score data (n = 726).

This study provides strong evidence that higher prenatal exposure to acrylamide through maternal diet during pregnancy is associated with reduced birth weight and head circumference. Prenatal exposure to acrylamide was evaluated using Hb adducts, which are well-established biomarkers for acrylamide exposure (Vikström et al. 2011). Birth weight decreased monotonically with increasing acrylamide and glycidamide adduct cord blood levels. These findings were consistent between countries and were shown in the full study population as well as in nonsmokers. Maternal consumption of acrylamide-rich foods was associated with higher levels of acrylamide and glycidamide Hb adducts in cord blood and with lower birth weight.

Prenatal exposure assessment using biomarkers is a key strength of the study. The measurement of Hb adduct levels in cord blood enabled a more accurate estimation of prenatal exposure to acrylamide compared with estimates based solely on dietary questionnaires or maternal Hb adduct levels. When cord blood Hb adduct levels are measured, variation related to transplacental exposure, uptake, and metabolism among children is taken into account. In line with our observation of higher Hb adduct levels in children from England compared with those of other study countries, higher levels have also been reported in nonsmoking British adults compared with other European adults (Vesper et al. 2008). Furthermore, our study population was large, and detailed information on maternal characteristics, including smoking and dietary habits during pregnancy, was collected in a manner that enabled us to evaluate potential sources of exposure contributing to in utero formation of Hb adducts and reduce potential biases through adjustment in a large subset of the study population.

In addition to dietary intake of acrylamide (Brantsaeter et al. 2008; Tran et al. 2010; Vikström et al. 2011), it is possible that acrylamide adducts were acting as a proxy marker for another dietary exposure or mix of exposures that were responsible for the associations observed, such as other Maillard reaction products that, like acrylamide, are formed during processing of food at high temperatures (Chaundhry et al. 2006; Tareke et al. 2002). Alternatively, acrylamide adducts may have been acting as a marker of a less healthy diet in general. Acrylamide may also be one of many contributors to the observed association. Adjusting for indicators of healthy and unhealthy eating habits (such as fruits and vegetables, fish, and soft drinks), maternal BMI, and indicators of socioeconomic status, did not substantially alter associations among the subset of the population with available data. Furthermore, foods that are generally considered to be part of a healthy diet, such as crisp bread and certain breakfast cereals, may contain high concentrations of acrylamide (EFSA 2010). Finally, Hb acrylamide and glycidamide adducts could also reflect exposure to tobacco smoke (von Stedingk et al. 2011). Monotonic dose–response associations of adduct levels with birth outcomes were observed in women who were nonsmokers in pregnancy, as well as in never-smokers, even after adjusting for passive smoking based on self-reporting or using ethylene oxide adducts as biomarkers of exposure to tobacco smoke. We cannot rule out uncontrolled confounding or the possibility that adduct levels are serving as a proxy marker for some other causal factor; however, given the consistency of our findings in different population subgroups and after adjustment for multiple potential confounders, it seems unlikely that prenatal exposure to tobacco smoke or other dietary compounds could fully explain the associations observed between acrylamide and birth weight.

A limitation of exposure assessment based on dietary information is the difficulty of evaluating exposure to toxic agents, such as acrylamide, for which concentrations may vary substantially among similar foods, depending on manufacturing or preparation methods (EFSA 2010; JECFA 2011; Manson et al. 2005; Tareke et al. 2002). Quantitative estimation of dietary intake is particularly complex in international studies. Nonetheless, intakes of key food types have been found to predict higher acrylamide Hb adduct levels in other study populations (Tran et al. 2010; Vikström et al. 2011) as well as in the present study population.

Decreases in offspring body weight following maternal acrylamide exposure during gestation have been consistently observed in mice and rats (El Sayyad et al. 2011; Tyl and Friedman 2003). A U.S. National Toxicology Program evaluation panel concluded that acrylamide causes decreased birth weight in rodents (Manson et al. 2005), although the mechanisms underlying the effects of acrylamide on birth weight are not understood. Acrylamide is classified as a probable carcinogen to humans on the basis of animal studies and genotoxicity (IARC 1994), and its genotoxicity is thought to be attributable largely to metabolic conversion to the genotoxic epoxide glycidamide (Rice 2005). Acrylamide Hb adduct levels in adults were associated with decreased serum insulin and reduced insulin resistance (Lin et al. 2009). In addition, oxidative stress causing increased production of reactive oxygen radicals and inflammation was reported in 14 healthy volunteers after ingestion of 160 g of fried potato crisps daily for 4 weeks, which correspond to a daily acrylamide exposure that is approximately three times higher than the currently calculated ingestion of approximately 50 µg/day in the Western diet (Naruszewicz et al. 2009). The ability of acrylamide to readily react with sulfhydryl and amino residues in proteins, including enzymes, receptors, and cytoskeletal proteins, can affect a multitude of cellular processes and has been suggested to form the basis of some of acrylamide’s toxic effects (IARC 1994) and may contribute to the associations with birth outcomes observed in our study population.

The potential public health implications of our findings are substantial. Increases in head circumference are an important indication of continued brain growth, and reduced birth head circumference has been associated with delayed neurodevelopment (Gale et al. 2006). Reduced birth weight is a risk factor for numerous adverse health effects early in life, and has been associated with multiple adverse outcomes later in life, such as reduced stature, increased incidence of cardiovascular disease, type 2 diabetes mellitus, and osteoporosis (Gluckman et al. 2008). The estimated difference in mean birth weight among children in the highest acrylamide-exposed quartile compared with children in the lowest quartile was around 100 g, consistent with the reduction in birth weight observed for children exposed in utero to maternal smoking (Li et al. 1993). Many commonly consumed foods—fried potatoes and related products, crisp bread, and coffee—contain high concentrations of acrylamide (EFSA 2010; JECFA 2011; Manson et al. 2005; Tareke et al. 2002). The amount of acrylamide in specific foods varies widely depending on precursor levels and processing, and, crucially, is amenable to appropriate interventions in the preparation of food.

In summary, this large population-based study provides the first epidemiological evidence of a significant association between prenatal exposure to acrylamide and reduced birth weight and head circumference. If confirmed in other studies, these findings provide evidence supporting the need for changes in food production and for providing clear public health advice to pregnant women to reduce their dietary intake of foods that may contain high concentrations of acrylamide.

Victoria J. Burley,¹ Ramon Carreras,² Vincenzo Fontana,³ Theo M. de Kok,⁴ Margaretha Haugen,⁵ Kari Hemminki,⁶ Micheline Kirsch-Volders,⁷ Antonis Koutis,⁸ Martinus Løvik,⁹ Patricia A. McKinney,¹⁰ Helle M. Meltzer,⁵ Renee Mijal,¹⁰ Elena Stagi,³ Simone G. J. van Breda,¹¹ Christopher P. Wild.¹⁰

Some of the values in Table 4 were incorrect in the manuscript originally published online. They have been corrected here.

1School of Food Science and Nutrition, University of Leeds, Leeds, United Kingdom; 2IMIM (Hospital del Mar Research Institute), Barcelona, Spain; 3Epidemiology, Biostatistics, and Clinical Trials, National Cancer Research Institute, Genoa, Italy; 4Department of Toxicogenomics, Maastricht University, Maastricht, the Netherlands; 5Department of Food Safety and Nutrition, Division of Environmental Medicine, Norwegian Institute of Public Health, Oslo, Norway; 6Division of Molecular Genetic Epidemiology, German Cancer Research Center, Heidelberg, Germany; 7Laboratory of Cell Genetics, Faculty of Science and Bio-engineering, Vrije Universiteit Brussel, Brussels, Belgium; 8Department of Social Medicine, Faculty of Medicine, University of Crete, Heraklion, Greece; 9Department of Environmental Immunology, Norwegian Institute of Public Health, Oslo, Norway; 10Leeds Institute of Genetics, Health and Therapeutics, University of Leeds, Leeds, United Kingdom; 11National School of Public Health, Athens, Greece.

We thank participants in the study and the doctors, nurses, midwives, and laboratory technicians who assisted with its conduct; and A. Espinosa, who helped with the statistical analysis.