Multiple mycotoxin exposure during pregnancy and risks of adverse birth outcomes: a prospective cohort study in rural Ethiopia

Introduction: Mycotoxin exposure during pregnancy has been associated with adverse birth outcomes in low- and middle-income countries. The evidence, however, is inconsistent and mainly limited to the assessment of a single mycotoxin. We assessed biomarkers of exposure to multiple mycotoxins during pregnancy and their associations with adverse birth outcomes in rural Ethiopia. Methods: We analyzed data from 579 pregnant women between 8 and 24 weeks of completed gestation enrolled in a prospective cohort study. Serum mycotoxin concentrations were determined using liquid chromatography coupled with tandem mass spectrometry. Multivariable linear probability models, adjusted for potential con- founding factors and multiple comparisons, were fitted to assess the associations between mycotoxin exposure and small for gestational age and preterm birth. We applied principal component analysis to reduce the dimensionality of biomarker data from several taxonomic mycotoxin groups. Results: All pregnant women were co-exposed to at least five mycotoxins, and one pregnant woman was co- exposed to 27 mycotoxins. Fumonisins (FB), i.e. , FB 2 , FB 3 , FB 1 , and tenuazonic acid were the most frequently identified mycotoxins in 98.8, 95.3, 93.3, and 81.4% of the samples respectively. Deoxynivalenol was detected in 38.7%, nivalenol in 50.1%, ochratoxin α in 67.9%, and zearalenone in 50.9% of the serum samples. After adjustment, we found no statistically significant (all P ≥ 0.05) associations between mycotoxin exposures and birth outcomes. Conclusions: Despite our study providing no evidence for relationships between mycotoxin biomarkers and adverse birth outcomes, our findings do indicate an extensive presence of multiple mycotoxin exposure among pregnant women. Public health policies and nutrition-sensitive interventions must ensure exposure to myco- toxins is reduced in rural Ethiopia. Principal component analysis was applied to generate PC factors explaining the variation in mycotoxin group data. Model 1 was multivariable-adjusted model for all other latent factors, whereas Model 2 was additionally adjusted for covariates of maternal age, height, weight, parity, depression, wealth tertiles, household food insecurity, maternal education, agro-ecological zone, and season.


Introduction
Mycotoxins are secondary metabolites produced by toxigenic fungi in plants. In sub-Saharan Africa, women and children are at higher risks of chronic (multiple) mycotoxin exposures, predominantly due to monotonous diets based on contaminated staple food crops (e.g., aflatoxins (AFs) and fumonisins (FBs) in rice and maize) (Andrews-Trevino et al., 2020;Kimanya et al., 2008;Gong et al., 2008). Fungal toxin contamination of food products might translate into detrimental acute or chronic health outcomes for consumers e.g., aflatoxin B 1 (AFB 1 ) is a cause of aflatoxicosis and a major risk factor for hepatocellular carcinoma (Wild and Gong, 2010;Kamala et al., 2018). Moreover, AFs are known to transverse the placental barrier, which might lead to embryonic or fetal exposure during critical developmental stages (Partanen et al., 2010;Stillerman et al., 2008). Mechanistic studies have indicated that in utero exposure to AFs might cause adverse birth outcomes by inducing environmental enteric dysfunction, upregulating proinflammatory cytokines, downregulating anti-inflammatory cytokines, and increasing the toxicity to maternal and fetal organs (Smith et al., 2012;. Numerous studies have indicated high prevalence of mycotoxin exposures during pregnancy in low-and middle-income countries (LMICs) (Piekkola et al., 2012;Leroy et al., 2015;Groopman et al., 2014). Consequently, there is heightened interest to quantify the effect of multiple mycotoxins on fetal growth restriction (Turner et al., 2007;Passarelli et al., 2020) and attributable post-natal child linear growth faltering (Gong et al., 2004;Leroy et al., 2018;Chen et al., 2018). Newborns with adverse pregnancy outcomes, such as low birthweight (LBW), small for gestational age (SGA), and preterm birth (PTB), have increased risks of morbidity and mortality during the neonatal and postneonatal periods, impaired growth and neurodevelopment, and increased health and development risks throughout their lifetime (Petrou, 2003;Katz et al., 2013;Christian et al., 2013). Moreover, Rasheed et al. estimated that AF-related stunting caused loss of between 3 and 36% of disability adjusted life years in low-income African countries (Rasheed et al., 2021).
Fumonisin B 1 (FB 1 ) might be an important risk or contributing factor for epigenetic dysfunction-associated diseases (Sugiyama et al., 2021), including esophageal cancer in humans (Shephard, 2011). Moreover, limited research has also revealed that chronic maternal exposure to FBs during early pregnancy is associated with an increased incidence of neural tube defects in their offspring (Missmer et al., 2006;Marasas et al., 2004). However, mechanistic evidence is still lacking. Zearalenone (ZEN) mycotoxin has been shown to have estrogenic properties on animals; however, the evidence that they may pose a risk to humans is limited (Kuiper-Goodman et al., 1987).
Recent studies have indicated that humans are more frequently exposed to multiple, rather than to a single mycotoxin (Martins et al., 2019;Heyndrickx et al., 2015), which has raised concerns about the potential combined effects of multiple mycotoxin exposure on human health. Nevertheless, most epidemiological studies assessing linkages between mycotoxins and adverse health outcomes have focused on the independent effects of single biomarkers (e.g., urinary aflatoxin M 1 (AFM 1 ) or FB 1 and serum aflatoxin B 1 -lysine (AFB 1 -lysine)) (Chen et al., 2018). However, to our knowledge, longitudinal studies evaluating the effects of multiple mycotoxin exposure, within taxonomic groups (e.g., AF, FB, ZEN, deoxynivalenol (DON)), on adverse birth outcomes are currently absent (Eze et al., 2018). Several classes of mycotoxins have been identified and characterized to date. These taxonomic groups are based on their association with human diseases.
In Ethiopia, 12-36% of neonates were born SGA, whereas 10% of children were born PTB in 2012 (Lee et al., 2017). Only a few studies have reported AF exposure among Ethiopian mothers (Eshete et al., 2021) and children (Ayelign et al., 2017) based on biomarkers, rather than the contamination of staple cereals (Getachew et al., 2018;Ayelign et al., 2018;Ssepuuya et al., 2018;Mesfin et al., 2021). Nevertheless, associations between multiple mycotoxin exposures during pregnancy and rates of adverse birth outcomes have yet to be documented. Using data from a prospective cohort study in rural Ethiopia, we quantified maternal mycotoxin exposures, in blood samples collected during pregnancy, and assessed the relationships with adverse birth outcomes in their offspring.

Study design and setting
We used data from the (ongoing) Butajira Nutrition, Mental Health, and Pregnancy (BUNMAP) cohort study collected between October 2017 and November 2020. The BUNMAP cohort was established under the Butajira Health and Demographic Surveillance Site (BHDSS), which consists of nine rural and one urban administrative sub-districts, representing the lowland, midland, and highland agro-ecological zones in the district. Khat (Catha edulis) and chili peppers are the key local cash crops, while maize, banana, and enset (Ensete ventricosun) are the main staples (Hassen et al., 2020).
The BUNMAP open cohort study planned to follow-up pregnant women and their newborns (up until 59 months of age) to evaluate the role of maternal nutrition, mycotoxin exposure, and mental health on prenatal and postnatal growth and development. Pregnant women were identified through active house-to-house surveillance in the BHDSS. All women aged 15-49 years who were between eight and 24 weeks of completed gestation and planning to deliver in the study area were enrolled in the study. Exclusion criteria included women with multifetal pregnancy, known pre-existing medical conditions, or a preceding pregnancy with complications (i.e., abortion, stillbirth, or neonatal death). For the current study, we analyzed data from 579 pregnant women from whom serum samples were collected at baseline. Birth outcomes were measured among 483 (83.4%) participants (Fig. 1).
The Institutional Review Board of Addis Ababa University, College of Health Sciences (099/17/SPH) approved the BUNMAP study protocol. Eligible mothers were asked to provide written informed consent of participation after an information session detailing the study objectives, voluntary participation, and rights to study withdrawal. Study participants who were anemic (hemoglobin < 11 g/dL) were referred to a local health center for follow-up. Our study is reported as per the Strengthening the Reporting of Observational Studies in Epidemiology-Nutritional Epidemiology (STROBE-nut) guideline (Lachat et al., 2016).

Mycotoxin exposure biomarkers: serum analysis
Potential multiple mycotoxin exposure was assessed from maternal serum according to an adapted methodology, which has been previously published (De Ruyck et al., 2020). Briefly, aliquot serum samples were acclimatized at room temperature in Eppendorf tubes and then vortexed. One-hundred and fifty µL of the serum sample was then spiked with 10 µL of internal standards and mixed with 150 µL liquid chromatography-mass spectrometry (LC-MS) grade acetonitrile (C 2 H 3 N) for protein precipitation. After centrifugation (4000 g, 10 min), 240 µL of the upper layer, the supernatant, was taken, and then evaporated to dryness under a gentle N 2 stream with Turbovap at 40 • C for 15 min. The residue obtained was reconstituted by vortexing in 150 µL of injection solvent (H 2 O/MeOH, 60/40, v/v), filtered (0.22 μm, PVDF, Durapore®, Cork, Ireland), and transferred to a UPLC vial upon LC-MS/MS analysis. The analysis of the samples was performed on a Waters® Acquity UPLC system coupled to a Quattro XEVO TQ-S mass spectrometer (Waters®, Manchester, UK). All instrumental parameters are detailed in a previous study (De Ruyck et al., 2020). We used MassLynx™ version 4.1 and QuanLynx® version 4.1 (Waters®, Manchester, UK) software to acquire and process relevant data.

Birth outcomes
Our dependent variables were SGA and PTB. PTB is defined as a child being born before 37 completed weeks of gestation (Howson et al., 2013), whereas SGA is defined as a birthweight less than the 10th centile for a specific completed gestational age by gender. SGA was calculated according to the international newborn size standards developed by the International Fetal and Newborn Growth Consortium for the 21st Century (INTERGROWTH-21st) (Villar et al., 2014).

Covariates
Covariates included in our multivariable-adjusted models were chosen a priori based on relevant literature (Kiserud et al., 2017;Tamirat et al., 2021). To construct a household wealth index across the BUNMAP cohort, we conducted a principal component analysis (PCA) using 26 wealth index constructs adapted from the Ethiopian Demographic and Health Survey (Central Statistical Agency (CSA) [Ethiopia] and ICF, 2016). Our models included the following potential confounders: maternal age (years), height (cm), weight (kg), education, parity (n), depression (yes; no), household wealth index (tertiles), and food security status (food secure; mildly food insecure; moderately food insecure; severely food insecure). Furthermore, we controlled our analyses for each agro-ecological zone and season as a fixed effect covariate (lowland; midland; highland).

Data collection
Study participants were invited to travel to the nearest health center for a comprehensive baseline assessment, which included a sociodemographic questionnaire, maternal anthropometry, hemoglobin measurement (g/dL), ultrasound examination, and the collection of maternal blood samples. A team of enumerators was given a ten-day training prior to conducting interviews among pregnant women.
Maternal depression was measured using the Patient Health Questionnaire-9 (PHQ-9) (Kroenke et al., 2001), whereas the Household Food Insecurity Access Scale (HFIAS) was applied to capture the access component of household food security (Swindale and Bilinsky, 2006). Maternal height and weight were measured using a Harpenden Pocket Stadiometer and a Tanita HD-314 digital scale, respectively. Hemoglobin concentrations were determined from finger-prick blood samples using a portable hemoglobin analyzer (HemoCue® Hb 301, Ä ngelholm, Sweden). Pregnant women with hemoglobin levels < 11 g/dL were considered anemic. Gestational age at baseline was estimated by an experienced sonographer using a portable diagnostic imaging and full-color, flow-mapping ultrasound system (SonoSite M-Turbo, FUJI-FILM SonoSite Inc., Bothell, WA 98021 USA) (Roro et al., 2019).
A trained phlebotomist collected a venous blood sample (5 mL) from eligible women. A temporary field laboratory in the study-area was set up to allow for immediate centrifugation of whole blood samples and preparation of the aliquots. The whole blood samples were taken according to a standardized operation protocol, performed by health-care personnel. Within one hour of collection, the samples were centrifuged, and the supernatant was transferred into the sterile cryovials using a polyethylene pipette. Serum samples were transported in cold boxes containing frozen gel packs (− 20 • C) to the Ethiopian Public Health Institute (EPHI) for laboratory analyses immediately after collection and stored at − 40 • C. Samples were shipped in dry ice to the Centre of Excellence in Mycotoxicology and Public Health, Faculty of Pharmaceutical Sciences, Ghent University, Belgium and stored at − 80 • C to laboratory analyses. Community-based field workers measured children's birth weights within 72 h of delivery to the nearest 0.01 kg using a digital baby scale (Tanita BD 585, Tokyo, Japan). Data collection was conducted electronically using the Open Data Kit software on Android tablets. Enumerators crosschecked the completeness of a questionnaire and first submitted it to the supervisor for confirmation. The supervisors checked the data quality before transferring to a central database at EPHI.

Statistical analysis
Descriptive statistics were presented as mean ± SD or median [interquartile range (IQR)] for continuous variables, and as frequencies and percentages for nominal variables. Exposure to a mycotoxin was estimated as the n (%) of positive samples. We followed a complete case analysis, where subjects without birth outcomes data collected and those without blood samples at baseline were not included in our analysis. We fitted linear probability models with robust variance estimators to estimate the associations between mycotoxin exposures and birth outcomes (i.e., SGA and PTB). Firstly, we evaluated the univariate associations between each mycotoxin exposure and birth outcomes. Secondly, multivariable-adjusted regression models were used to assess the (independent) associations between mycotoxin exposures and birth outcomes. To avoid over adjustment and multi-collinearity across the individual mycotoxin variables, a data reduction approach was applied to summarize mycotoxins under their predefined taxonomic classification. For this purpose, we applied PCA with varimax rotation to generate latent factors explaining the variance within each taxon of mycotoxins (O'Rourke and Hatcher, 2013). After identifying the factors which accounted for most of the variance, based on scree plot and eigenvalues (≥1), these latent factors were used to build the models. Thus, Model 1 β-coefficients were adjusted for other principal components (PC) only, whereas Model 2 β-coefficients were additionally adjusted for potential confounding factors. To avoid type-I error inflation due to multiple hypotheses testing, we used the Benjamini-Hochberg method to control the false discovery rate. False discovery rate controlling procedures have a less stringent control over false discovery compared to familywise error rate procedures such as the Bonferroni correction (Benjamini and Hochberg, 1995).
Data management and statistical analysis was performed using Stata version 17.0 (StataCorp LLC, College Station, TX, USA). A two-sided significance level of P < 0.05 was applied for all analyses.

Results
A flow chart of the study is presented in Fig. 1. From the 776 eligible women enrolled in the BUNMAP cohort, 196 (25.3%) mothers were excluded due to missing blood samples. Subsequently, we excluded an additional 96 (12.4%) women due to missing birth weight data, occurrence of an adverse birth outcome, or a multifetal pregnancy. We followed-up the remaining 483 mothers for a median duration of 22 weeks.

Maternal characteristics and pregnancy outcomes
Baseline characteristics of 579 pregnant women are shown in Table 1. On average, pregnant women were 25.8 ± 4.58 years old at enrollment and their median (IQR) parity was 2.00 (1.00, 4.00). Approximately 48% of mothers were from a food insecure household. Moreover, maternal BMI was on average 21.5 ± 2.85 kg/m 2 , whereas gestational age at enrollment was 16.8 ± 4.58 weeks. About one-third of the participants received no formal education. Baseline hemoglobin concentration was 13.1 ± 1.44 g/dL and 5.70% of pregnant women were anemic. Gestational age at birth was 38.9 ± 2.44 weeks and children's birth weight was 2935 ± 459 g. Notable is that 27.2% (n = 131) of the newborns were SGA, 11.6% (n = 56) were LBW, and 19.1% (n = 92) were born PTB.

Relationships between maternal mycotoxin exposures and adverse birth outcomes
The univariate associations between each mycotoxin biomarker and children's birth outcomes are presented in Table 3. Mycotoxins that were significantly associated with a higher rate of SGA were aflatoxin G 1 (AFG 1 ), HT-2, and TeA. AFB 1 , beta zearalanol (β-ZAL), and OTα were associated with a lower PTB rate. However, none of the observed relationships were statistically significant (all P ≥ 0.05) after adjustment for multiple comparisons.
Our PCA analysis indicated that the AFs group was represented by two PC factors explaining 49% of the total variance. Aflatoxin B 2 (AFB 2 ), AFG 2 , AFM 1 , and AFB 1 -lysine represented the first PC factor, whereas AFB 1 and AFG 1 were indicative of the second PC factor. FBs group was characterized by one PC factor representing FB 1 , FB 2 , FB 3 , and hydrolyzed fumonisin B 1 (HFB 1 ), which explained 72% of the total variance. The trichothecenes group was represented by four PC factors, which explained more than 67% of the total variance. The first PC factor represented DON, 3-acetyldeoxynivalenol (3-ADON), deepoxydeoxynivalenol (DOM), NIV, and neosolaniol (NEO), the second represented T-2 toxin (T-2) and HT-2, the third only DON3G, and the fourth fusarenon (FUS-X) and diacetoxyscirpenol (DAS). ZEN group was characterized by one PC factor representing ZEN, alpha zearalenol (α-ZEL), β-ZAL, and zearalanone (ZAN), which explained 50% of the total variance. Similarly, OTA group was characterized by one PC factor explaining 52% of the total variance, which represented OTA and OTα. Lastly, Alternaria group was represented by two PC factors explaining 75% of the total variance. Our first factor represented AOH and alternariol monomethyl ether (AME), whereas the second represented TeA.

Discussion
To our knowledge, this is the first epidemiological study that comprehensively assessed maternal multiple mycotoxin exposures in serum (i.e. 33 biomarkers) during pregnancy and associations with adverse birth outcomes. Our findings show the presence of multiple mycotoxin exposure among mothers living in resource-poor settings in Ethiopia, and indicate a high prevalence of mycotoxin co-exposure. After adjustment for potential confounders (including other mycotoxin PC factors) and testing of multiple hypotheses, our findings provide no evidence for statistically significant associations between maternal mycotoxin exposure and SGA or PTB rates.
In the BUNMAP cohort, FB 1 , FB 2 , and FB 3 were the most frequently detected biomarkers in pregnant women's serum samples (all > 95%). Our findings are similar to prevalence estimates of urinary FB 1 among children in two Tanzanian studies (i.e., 80 and 98% positive samples, respectively) (Chen et al., 2018;Shirima et al., 2015). Research has shown that FB 1 might be an important risk factor or contributing factor for epigenetic dysfunction-associated diseases (Sugiyama et al., 2021), including esophageal cancer (Shephard, 2011) and neural tube defects in animals and humans (Missmer et al., 2006;Marasas et al., 2004). However, mechanistic evidence is still lacking. Furthermore, we report that over 85% of mothers' serum samples had detectable levels of AFB 1lysine, which is in concurrence with studies among pregnant women in Tanzania (92%) (Passarelli et al., 2020), Nepal (94%) (Andrews-Trevino et al., 2019), The Gambia (Turner et al., 2007) and Uganda (both 100%) (Lauer et al., 2019), and young children in Tanzania (72%) (Chen et al., 2018), Mexico , and Kenya (both ~100%) (Hoffmann et al., 2018). AF exposure is hypothesized to result in damage of the intestinal mucosa and hence nutrient malabsorption and increased gut permeability, immunomodulation, DNA methylation, and alteration in the insulin-like growth factor axis caused by liver damage (Abu et al., 2008;Hernandez-Vargas et al., 2015;Xu et al., 2021). Lastly, our findings on widespread presence of emerging Alternaria toxins is of concern, as this taxonomic group has been shown to induce fetotoxic and teratogenic effects in animals (Pollock et al., 1982;Fraeyman et al., 2017). In particular, TeA (84.1% positive samples) is considered the most toxic, and many in vivo studies have demonstrated that it causes severe pathophysiological effects, such as intestinal multi-hemorrhages and impaired liver and kidney functions (Fraeyman et al., 2017). In contrast to our results, a mycotoxin exposure assessment by De Ruyck et al. did not detect any serum Alternaria toxins (i.e., TeA, AME, and AOH) among adults from five European countries (De Ruyck et al., 2020). In our study population, all 33 mycotoxin biomarkers analyzed were detected in serum samples, with a co-exposure of at least five mycotoxins. Few studies have reported on co-exposure to mycotoxins, in part, due to the limited number of validated mycotoxin biomarkers (Vidal et al., 2018). Nevertheless, similar to a biomonitoring study among adults in China (Huang et al., 2021), we report a high frequency of mycotoxin coexposure in pregnant women in Ethiopia.
Our results indicate relatively low mycotoxin biomarker concentrations in Ethiopia as compared to e.g., AF levels among pregnant women in South Asia (Groopman et al., 2014) or The Gambia (Turner et al.,

Table 2
Mycotoxin exposures in serum samples of pregnant women (n = 579). 2007). However, there are currently no thresholds of safe human mycotoxin exposure in biological specimens (De Ruyck et al., 2020). Moreover, the cumulative effect of co-exposure to multiple mycotoxins at low concentrations remains unknown, due to a lack of robust methodological approaches to analyze the combined, potentially highly correlated, effect of mycotoxin biomarker data. Nonetheless, preliminary evidence from in vitro models suggests that while single mycotoxin exposures at low doses are non-toxic, various combinations of toxins at equal doses can give rise to complementary or even synergistic toxicity (Wan et al., 2013). Although there remains limited mycotoxin toxicity and toxico-kinetic modelling data or validated biomarkers of exposure in biological matrices (e.g., urine, blood, hair), large human biomonitoring initiatives, such as the HBM4EU-program (htt ps://www.hbm4eu.eu/), aim to assess actual mycotoxin exposure in distinct populations and their potential adverse health outcomes. Our findings from bivariate regression models are, at least qualitatively, comparable to (crude) associations estimated between prenatal AFB 1 -lysine exposure and pregnancy outcomes in five prospective cohorts. Our results indicated no statistically significant associations between mothers' AFB 1 -lysine and SGA or PTB rates, although serum concentrations were relatively low in the BUNMAP study. In parallel, Passarelli et al. (2020) and Andrews-Trevino et al. (2019) showed no statistical relationships (i.e., unadjusted estimates) between maternal AF exposure and SGA, PTB and LBW rates in Tanzania and Nepal, respectively. Furthermore, Turner et al. reported that higher maternal AF exposure was not associated with children's birth weight or length in The Gambia (Turner et al., 2007), whereas De Vries et al. showed significantly higher birth weight among boys (but not girls) born to AFnegative mothers in Kenya (De Vries et al., 1989). Lauer et al. reported significant inverse associations between AFB 1 -lysine and birth weight and head circumference, but not birth length or completed weeks of gestation in Uganda (Lauer et al., 2019).
Our multivariable-adjusted models; which included other mycotoxin PC factors, a priori defined confounders, or both, showed inconsistent directions, and thus likely spurious associations between mycotoxin biomarkers, across and within taxonomic groups, and SGA or PTB. Moreover, all observed relationships were non-significant after adjusting for multiple testing of hypotheses.
Epidemiological studies on mycotoxin and children's birth outcomes, which have mainly focused on AFs and provide mixed evidence, have important variations in study design, analytical techniques, biological specimens collected, concentrations of mycotoxin biomarkers (i. e., exposure), assessment methods of gestational age, timing of pregnancy exposure, and (measured) confounder adjustment (Tesfamariam et al., 2020).

Strengths and limitations of the study
Most mycotoxin exposure studies have considered only a single (Mahfuz et al., 2021) or a limited number of mycotoxins (Chen et al., 2018) e.g., urinary FB 1 or serum AFB 1 -lysine, or one mycotoxin group  when assessing relationships with adverse birth outcomes. Our quantification of a suite of serum biomarkers should be regarded as both a study strength and limitation. Mycotoxin biomarkers have advantages over foodstuff (De Boevre et al., 2013) or dietary exposure assessments as they provide more objective data on mycotoxins exposure, and correct for the heterogeneous distribution of mycotoxins in food (De Ruyck et al., 2020). However, only urinary AFM 1 , aflatoxin-N7-guanine, and DON-glucuronides and blood AFB 1 -lysine have been validated in humans (Vidal et al., 2018). Hence, our findings using non-validated biomarkers (e.g., serum FBs) should be interpreted with caution. Although our study has a wide scope of the most adequate mycotoxins and their metabolites, a significant underestimation of mycotoxin exposure is plausible, especially for heavily metabolized mycotoxins (e.g., AFB 1 -lysine was found in 85% of the serum samples, while AFB 1 was found in only 7.43%). To date, multiple metabolites are still not commercially available enabling quantitative measurements, and validated biomarkers for some mycotoxins have not been identified so far. However, the use of liquid chromatography-high-resolution mass spectrometry (LC-HRMS) in untargeted mode has allowed the identification of metabolites of mycotoxins (Lauwers et al., 2019). Nevertheless, considering the lack of commercial standards, LC-HRMS provides qualitative rather than quantitative benefits to mycotoxins analysis. In the present study, our focus remained on quantification. Thus, further research is required to harness the combination of LC-MS/MS and LC-HRMS, which enables detecting a broad range of mycotoxins, i.e., the

Fig. 2.
Histogram of mycotoxins co-exposure among 579 pregnant women in Ethiopia. Exposure to a mycotoxin was defined as a serum concentration above the limit of detection. parent compounds and their predominant metabolite(s) to accurately estimate mycotoxins exposure (Vidal et al., 2018). On the other hand, compared to previous studies that applied conventional analytical methods, our LC-MS/MS is a highly sensitive and specific analytical technique that can precisely detect low mycotoxins concentrations; therefore, high exposure levels reported can be attributed to the use of a sensitive analytical method.
Furthermore, our prospective study design allows for stronger causal interpretation, as the concentrations of mycotoxin biomarkers were assessed simultaneously and prior to pregnancy outcomes. Moreover, we assessed gestation by an objective ultrasound, rather than the more error-prone recall (e.g., irregular menses, variations in length of menstrual cycle) of a mother's last normal menstrual period (Savitz et al., 2002). Our study also adjusted for potential confounders, which included other mycotoxin PC factors, and sampled across distinct agroecological zones and seasons. However, our analyses were unable to account for potential (unmeasured) residual confounding, including dietary variables, such as dietary diversity , food quantities or total energy consumed . Lastly, mycotoxin exposures were only measured at a single time point during pregnancy. However, the effects of mycotoxins and their physicochemical properties are not well understood in relation to the timing of exposure. Mycotoxin concentrations and their effects may vary between first, second, and third trimesters; hence, studies collecting repeated measures (i.e., bio-specimens) are warranted to identify specific windows of fetal vulnerability.

Conclusions
In the BUNMAP prospective cohort study, pregnant women were exposed to multiple mycotoxins, for which evidence of human health effect is currently mixed or missing. Nevertheless, public health policies (e.g., food safety regulations and their enforcement) and nutritionsensitive interventions must ensure exposure to mycotoxins is reduced. Despite the absence of statistically significant associations with adverse birth outcomes, our findings do indicate widespread prevalence of mycotoxin co-exposure among pregnant women living in rural Ethiopia. Mycotoxins not only pose major acute and chronic risks to both human and animal health, but also affect food and nutrition security by reducing access to safe and healthy food.

Funding
This work was supported by VLIR-UOS Network program (ET2017NET039A103), Ghent University (MYTOX-SOUTH® consortium), and NUFFIC (NICHE/ETH/179) Netherlands Initiative for Capacity Development in Higher Education. This work was also funded by Addis Ababa University's thematic research grant. The study was conducted in the context of a collaboration of four public universities in Ethiopia coordinated by Jimma University (JU) and five Flemish universities coordinated by Ghent University (UGent). MYTOX-SOUTH® offers research and expertise that deals with the occurrence of mycotoxins and their effect on human health. MDB is supported by the 1 β coefficients (95% CIs) and P-values for the associations between mycotoxin concentrations and birth outcomes estimated using linear probability models with robust variance estimation. P-values were non-significant (P ≥ 0.05) after adjustment for multiple comparisons using the Benjamini-Hochberg method.
European Research Council (ERC) under the European Union's Horizon 2020 research and innovation program (grant agreement No 946192, HUMYCO). The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the manuscript.