Shaping the risk for late-life neurodegenerative disease: A systematic review on prenatal risk factors for Alzheimer’s disease-related volumetric brain biomarkers

Environmental exposures including toxins and nutrition may hamper the developing brain in utero, limiting the brain's reserve capacity and increasing the risk for Alzheimer's disease (AD). The purpose of this systematic review is to summarize all currently available evidence for the association between prenatal exposures and AD-related volumetric brain biomarkers. We systematically searched MEDLINE and Embase for studies in humans reporting on associations between prenatal exposure(s) and AD-related volumetric brain biomarkers, including whole brain volume (WBV), hippocampal volume (HV) and/or temporal lobe volume (TLV) measured with structural magnetic resonance imaging (PROSPERO; CRD42020169317). Risk of bias was assessed using the Newcastle Ottawa Scale. We identified 79 eligible studies (search date: August 30th, 2020; Ntotal=24,784; median age 10.7 years) reporting on WBV (N=38), HV (N=63) and/or TLV (N=5) in exposure categories alcohol (N=30), smoking (N=7), illicit drugs (N=14), mental health problems (N=7), diet (N=8), disease, treatment and physiology (N=10), infections (N=6) and environmental exposures (N=3). Overall risk of bias was low. Prenatal exposure to alcohol, opioids, cocaine, nutrient shortage, placental dysfunction and maternal anemia was associated with smaller brain volumes. We conclude that the prenatal environment is important in shaping the risk for late-life neurodegenerative disease.


Introduction
The fetal brain grows and develops at a remarkable speed. At three weeks post-conception, primitive cerebral hemispheres have already developed. By mid-gestation, the fetal brain has largely achieved the adult neuronal number (Dobbing and Sands, 1973;Prayer et al., 2006). As a result of this exceptional growth rate, the prenatal period is a critical period during which the developing brain is especially vulnerable to adverse exposures (Whalley et al., 2006).
Adverse exposures during prenatal development may impact the brain, hampering developmental processes and preventing it from developing to its full potential. For instance, prenatal exposure to alcohol or tobacco restricts fetal brain growth, mostly observed as reduced whole brain and regional brain volumes (e.g. Banderali et al., 2015;Caputo et al., 2016;Ekblad et al., 2015;Popova et al., 2021). Exposure to adverse prenatal circumstances is also negatively associated with cognitive functioning in (early) childhood. For example, maternal cannabis, cocaine or alcohol use during pregnancy has been associated with deficits in cognitive functioning and increased risk for psychopathology or substance use disorders in childhood (Grant et al., 2018;Paul et al., 2021;Singer et al., 2018).
There is some evidence that these harmful effects of prenatal exposures on brain structure and functioning last throughout life. Exposure to prenatal undernutrition was associated with poorer cognitive performance at the age of 58 (de Rooij et al., 2010). In addition, sex-specific effects of undernutrition during early gestation on brain volumes at the age of 68 were reported, demonstrating smaller volumes in exposed men (de Rooij et al., 2016). Thus, the impact of the prenatal environment on brain structure and cognitive functioning appears to be present in early childhood and may potentially last throughout life, although the number of studies on long-term brain-related outcomes of adverse prenatal exposures measured in late-life is limited.
Adverse prenatal circumstances may also be associated with an increased risk of dementia in late life (Borenstein et al., 2006;Seifan et al., 2015). Among numerous potential pathways, the increased risk of dementia could result from hampered development of the fetal brain following a suboptimal prenatal environment, limiting the reserve capacity of the adult brain. Brain reserve capacity can be defined as a buffer determined by neural factors such as brain size and the number of neurons and synapses (Borenstein et al., 2006). As the brain ages, neurodegeneration and vascular damage can accumulate, which may be part of normal ageing, or pathological, such as observed in Alzheimer's disease (AD). As neurodegeneration progresses, the brain reserve may determine whether an individual experiences symptoms of cognitive decline. At the same level of neurodegeneration, an individual with a large brain reserve may continue to function normally whereas someone with a limited brain reserve may reach the threshold of cognitive dysfunction earlier. Thereby, acting as a structural buffer, brain reserve can compensate for some of the neurodegeneration associated with AD in late life, alleviating the effects of initial neurodegeneration on cognitive functioning (Borenstein et al., 2006;de Rooij, 2022;Stern, 2012). Thus, through a limited development of brain reserve, adverse prenatal exposures may result in an increased risk of developing AD in late life.
AD is a neurodegenerative disease marked by amyloid-beta plaques, tau tangles and a general pattern of brain atrophy (Ausó et al., 2020). Magnetic resonance imaging (MRI) studies can provide markers of AD-related neurodegeneration (Pini et al., 2016). Generally, MRI studies of AD report reduced whole brain volume (WBV), temporal lobe volume (TLV) and hippocampal volume (HV) as established neurodegenerative volumetric biomarkers for AD (Franke et al., 2010;Frisoni et al., 2010;Hane et al., 2017;Pini et al., 2016;Wardlaw et al., 2013). Reduced HV is seen early in the neurodegenerative process (Pini et al., 2016), as is temporal lobe atrophy. Further progression of AD neurodegeneration is associated with widespread cortical atrophy across the brain (Pini et al., 2016). Both hippocampal and whole-brain atrophy rates are used as markers for neurodegenerative progression of AD (Frisoni et al., 2010). Hampered development of the HV, TLV and WBV as a result of adverse prenatal circumstances may lead to a reduced brain reserve to buffer AD-related neurodegeneration in late-life (Borenstein et al., 2006).
All in all, the prenatal environment may play an essential role in determining the risk for developing AD by impacting brain reserve. There is, however, no overview of the available body of evidence on prenatal exposures and their impact on structural brain measures that have been associated with an increased risk for developing AD in late life. The purpose of this systematic review is to summarize all currently available evidence for the association between prenatal exposures (including prenatal illicit drug exposure, alcohol exposure, maternal stress and environmental exposures) and HV, TLV and WBV in humans. Thereby, we aim to address the following research question: 'Is there an association between prenatal risk factors and volumetric MRI neuroimaging biomarkers for AD?'. The present systematic review is essential to gain insight into the extent to which the prenatal environment may shape the risk for late-life neurodegenerative disease through brain reserve.

Methods
We registered the protocol for this systematic review at the International Prospective Register of Systematic Reviews (PROSPERO; CRD42020169317) and followed PRISMA reporting guidelines for systematic reviews (Page et al., 2021;Supplement).

Search strategy
An information specialist (JL) searched OVID Medline and EMBASE from inception to August 30th, 2020. MESH terms and text words for 1. general prenatal terms, including antenatal and fetus, or specific prenatal exposures were combined with 2. terms for structural MRI brain biomarkers. We excluded conference abstracts and reviews, but did not apply language or date restrictions. The full search strategy is available in supplement B. The bibliographic records were imported and deduplicated using EndNote. We cross-checked reference lists and citing articles of identified relevant papers in Web of Science and adapted the search in case of additional relevant studies.

Eligibility criteria
We included peer-reviewed and published human cohort, crosssectional and case-control studies examining prenatal factors in association with predefined neuroimaging biomarkers related to sporadic lateonset AD. No restrictions were made for age at outcome measurement. We defined a prenatal factor as any exposure that occurs during pregnancy. All direct, specific exposures were of interest, indirect measures of the prenatal environment such as birth weight or other birth characteristics were not included. Studies focusing on premature birth as the only factor of interest were excluded, since premature birth can be the result of a large variety of potential exposures throughout pregnancy. Studies investigating specific prenatal exposures that may result in premature birth were included. We defined our outcomes of interest as WBV, TLV and/or HV as measured by structural MRI scans (Franke et al., 2010;Frisoni et al., 2010;Hane et al., 2017;Pini et al., 2016;Wardlaw et al., 2013).
In the original review protocol, additional vascular variables were listed as outcomes of interest. However, these outcomes were rarely identified in relation to direct prenatal exposures. Therefore, we additionally excluded the studies reporting on these outcomes to reduce the heterogeneity of the included studies.
We excluded studies involving specific clinical populations including those with Familial early-onset AD, vascular dementia, Korsakoff syndrome and Down's syndrome. Studies focusing on a clinical population (e.g. patients with congenital heart disease) were included only if they studied the association between the exposure and outcome of interest separately in the group of healthy controls. In this case, the results from the healthy control group were extracted. We restricted on analysis methods by excluding voxel-based morphometry studies, since this analysis approach is significantly different from other analysis methods, which makes the comparison of results between these studies difficult.

Study selection and data extraction
All screening steps were performed by two authors independently, blinded for the other's decisions. AB and AMW independently screened the title and abstract of retrieved papers using Rayyan software (Ouzzani et al., 2016). Discrepancies were discussed and, if needed, resolved by a third reviewer (SdR). Two pairs of authors screened full texts (AB and YV, AMW and MvdH). The same pairs of authors performed data extraction, where the data was extracted by one reviewer and checked by the other. All discrepancies were discussed in pairs, and discussed with all four reviewers if needed. In case of considerable overlap in study samples, we included the study with the largest number of participants. We extracted the following data items: study aim, design, year of publication, cohort (if applicable), location, population, number of participants (per group), participant age at outcome assessment, exposure(s), timing of exposure (if applicable), control conditions, exposure assessment, relevant outcome(s), outcome volume, outcome assessment methods, scanner details, analysis details, confounders/covariates adjusted for in the analyses, the statistical relationship between exposure and outcome(s) (unadjusted and adjusted for confounders) and sub-group results. Template study selection and data extraction forms can be requested from the corresponding author.

Risk of bias assessment
The same author pairs performed risk of bias assessment using the Newcastle Ottawa Scale (NOS) for assessing the quality of nonrandomized studies in duplicate, blinded to the other's decisions (Wells et al., 2000). Disagreements between the authors were discussed in pairs and resolved with all four reviewers if needed. Prospective cohort studies were assessed using the NOS for cohorts, cross-sectional studies were assessed using the adapted NOS for cross-sectional studies (Alshabanat et al., 2015). When a cohort study had no prospective elements, we used the cross-sectional NOS as well since the items in this checklist were better suited for this study design. We removed the item "Demonstration that outcome of interest was not present at start of study" from the Cohort checklist since this question was not suitable for the studies included in our review given the nature of prenatal exposures. Thereby, both checklists had a maximum total score of 8.

Data synthesis
The results are presented in the form of a structured narrative synthesis, with the studies categorized on exposure and outcome. Outcome effect measures were reported as brain volumes, effect size (%difference) and statistical significance. In general, statistical significance was set at p < 0.05. We did not perform meta-analyses given the large heterogeneity in study methodology and age at outcome assessment.

Results
Of 1696 unique studies, 84 were identified as eligible. Seven studies were subsequently excluded based on overlap in study participants Biffen et al., 2017;Coles et al., 2011;Fryer et al., 2012;Gross et al., 2018;Robey et al., 2014;Wu et al., 2020a). Two additional studies were identified through citation searching, resulting in 79 included studies in the final review (Fig. 1). These studies had a total of 24,784 participants from 14 countries.
A large proportion of included studies investigated prenatal exposure to alcohol (N = 30). Other exposures included illicit drugs (N = 14), smoking (N = 7), diet (N = 8), environmental exposures (N = 3), maternal disease, treatment and physiology (N = 10) with the subcategories infections (N = 6) and mental health problems (N = 7). For numerous studies, there was an overlap between multiple exposures, mostly concerning exposures to alcohol, tobacco and illicit drugs. Prenatal illicit drug exposure studies often evaluated multi-drug exposures and mental health studies mostly evaluated a combination of prenatal stress, anxiety and depression exposure.
The majority of studies included WBV (N = 38) and/or HV (N = 63) as an outcome, with only a few (N = 5) reporting TLV. Age at outcome assessment ranged between 22 weeks gestational age (GA) to 67 years postnatally, although most studies assessed the outcome in childhood (median age 10.7 years). An interactive bubble map of the evidence according to exposure category, outcome and direction of effect created using EPPI reviewer software is available as a supplementary file ( Fig. S1; Thomas et al., 2020).

Risk of bias assessment
An overview of risk of bias score per exposure category and study design is provided in Fig. 2 Supplementary  Table 3). Studies had a mean score of 5.8/8 on the NOS. With a maximum score of 8 on the NOS, 1 study had a score of 2/8 (1%), 3 studies were scored with a 3/8 (4%), 5 with a 4/8 (6%), 24 with a 5/8 (30%), 16 with a 6/8 (20%), 23 with a 7/8 (29%) and 7 with 8/8 (9%). Especially studies in the categories of environmental exposures, diet and illicit drugs had satisfactory scores on the NOS. In general, studies scored poorly on the description of response rate and characteristics of nonrespondents, follow-up adequacy and sample size justification.

.(detailed information in
For specific exposure categories, studies in the infections category were often hampered by an inadequate representativeness of the study sample. Both infection and alcohol exposure studies had a notable risk of bias in selection of the unexposed cohort. Smoking and mental health studies scored poorly on the ascertainment of exposure. Mental health studies often did not report the assessment of outcome. Maternal disease, treatment and physiology studies and infection studies had poor comparability of the cohorts.
Prenatal exposure to alcohol had inconsistent associations with TLV, with exposed individuals having a 16.4% smaller TLV at age 13 (Sowell et al., 2002), but no statistically significant effect was observed at age 12 and 11-15 (No volumes reported; Archibald et al., 2001;Treit et al., 2016).

Illicit drugs
We identified thirteen studies investigating the association between prenatal illicit drug exposure and WBV (N = 8) and/or HV (N = 10) ( Table 1, Supplementary Table 1 and 2). No studies included TLV as an outcome. Individuals prenatally exposed to opioids had a 3.0-16.0% smaller WBV compared to unexposed individuals at age 10-11, 12 and 19 (Nygaard et al., 2018;Sirnes et al., 2017;Walhovd et al., 2007) or to the population mean (41 weeks, Yuan et al., 2014). This was statistically significant in three studies. Sirnes et al. reported no significant association, although exposed individuals had a 3.7% smaller WBV (Sirnes et al., 2017). Of note, there was overlap in the study samples of Walhovd et al. and Nygaard et al., although the outcome was assessed at different ages (Nygaard et al., 2018;Walhovd et al., 2007). Smaller WBV was reported after prenatal exposure to cocaine, which was statistically significant at age 12 (Rivkin et al., 2008) and not significant at age 15 (Roussotte et al., 2012a). No statistically significant associations between prenatal cannabis (7 years, El Marroun et al., 2016), marijuana (Rivkin et al., 2008) or methamphetamine (7 years, Chang et al., 2004) and WBV were reported, despite the 5.9% smaller WBV reported by Rivkin et al. and 3.1% smaller WBV reported by Chang et al. in exposed children.
Studies reported a 4.4-7.0% smaller HV in individuals prenatally exposed to opioids compared to unexposed individuals at age 10, 12 and 19 (Nygaard et al., 2018;Sirnes et al., 2017;Walhovd et al., 2007). None of these effects was statistically significant. Prenatal exposure to cocaine had uncertain results on HV at age 9, 13 and 15, with studies reporting mixed effects (effect size range − 1.3% to +6.2%) with no statistically significant associations (Akyuz et al., 2014;Liu et al., 2013;Roussotte et al., 2012a). Of note, Akyuz et al. and Liu et al. included participants from the same cohort. Riggins et al. studied heroin and/or cocaine exposure, and reported a significantly larger left (+6.2%) and right (+5.3%) HV in exposed children (14 years, Riggins et al., 2012). Akyuz and colleagues additionally studied the growth of the HV between age 9 and 14 years, and reported a lower percent HV growth after prenatal cocaine exposure (1% versus 9% growth). Prenatal methamphetamine exposure had inconsistent results on HV, with a significantly smaller left (− 20.5%) and right (− 19.5%) HV at age 7, but no significant effects at 42 weeks GA and age 4 (Chang et al., 2004;Derauf et al., 2012;Warton et al., 2018). Warton et al. did report a larger HV after prenatal methamphetamine exposure (left +3.3% and right +2.5%), and Derauf et al. a smaller HV (− 2.6%), although both effects were not statistically significant.

Smoking
We included seven studies investigating an association between maternal smoking and WBV (N = 4) and/or HV (N = 4) ( Table 1, Supplementary Table 1 and 2). None of these studies included temporal lobe volume (TLV) as an outcome. Individuals prenatally exposed to smoking had a smaller WBV compared to prenatally unexposed individuals. This was statistically significant in three studies (GA 24-35 weeks, Anblagan et al., 2013;7 years, El Marroun et al., 2014;12 years, Rivkin et al., 2008). For the fourth study performed by De Zeeuw and colleagues, the reported average WBV was smaller for exposed offspring compared to unexposed offspring, but the reporting on statistical significance was unclear (11 years, de Zeeuw et al., 2012). This study also had the highest risk of bias.
No studies reported a statistically significant association between maternal smoking and offspring HV. Of note, exposed children did have a smaller HV in the two studies that reported brain volumes and three studies corrected for ICV in their analysis, thereby potentially correcting for developmental effects (4 years, Derauf et al., 2012;7

Diet
Seven cohort studies and one cross-sectional study investigated maternal diet in relation to WBV (N = 6) and/or HV (N = 6) ( Table 1,  Supplementary Table 1 and 2). No studies reported on the relationship between maternal diet and TLV.
Prenatal exposure to the Dutch famine had inconsistent results on WBV, with exposed individuals having a 6.7% larger WBV at age 51 (Hulshoff Pol et al., 2000), but a statistically significant smaller WBV (− 2.4% to − 7.3%) at age 67 (de Rooij et al., 2016). The smaller WBV at age 67 was mainly driven by a smaller WBV in exposed men. Of note, the effect reported by De Rooij et al. was only observed without correction for ICV, and Hulshoff Pol et al. reported results corrected for ICV. Therefore, it is unknown whether an ICV-dependent effect may have been present at age 51. Also, the number of participants in the analysis performed by Hulshoff Pol et al. was very small, limiting the power of this study to detect potentially smaller effect sizes. Studies reported no significant associations between maternal B12 or homocysteine levels (7 years, Ars et al., 2019) or prenatal LC-PUFA or 5-MTHF supplementation and WBV (10 years, Catena et al., 2019). Ars et al. did report a lower WBV in children exposed to low prenatal maternal folate levels compared to children exposed to normal prenatal maternal folate levels. Ogundipe and colleagues reported no statistically significant effect of essential brain-specific fatty acids supplementation across all participants (group difference range +1.9 to +3.1%). They did report a higher WBV in men after supplementation, pointing at a potentially differential sex sensitivity (0-4 weeks, Ogundipe et al., 2018). Lastly, Zou et al. reported a smaller WBV after exposure to continuous prenatal vitamin D insufficiency (10 years, Zou et al., 2020).
Alves et al. found a significant association between maternal prepregnancy BMI and HV, with a significant negative correlation between maternal prepregnancy BMI and HV in boys but not girls (8 years, Alves et al., 2020). Morton et al. reported a positive correlation between prenatal maternal omega-3 fatty acid intake reported by the mother and HV (27 days, Morton et al., 2020). This correlation was no longer significant after correcting for multiple testing and reporting of these results was limited. No statistically significant associations with HV were reported for low prenatal folate levels, B12 and homocysteine levels (7 years, Ars et al., 2019), prenatal LC-PUFA or 5-MTHF supplementation (10 years, Catena et al., 2019) or prenatal vitamin D levels (10 years, Zou et al., 2020). Furthermore, no significant effect of essential brain specific fatty acids supplementation was reported, although the supplementation group had a 1.9% larger HV (0-4 weeks, Ogundipe et al., 2018).

Environmental
Three studies investigated WBV (N = 2) and/or HV (N = 3) in relation to prenatal environmental exposures (Table 1, Supplementary Table  1 and 2). No studies included TLV as an outcome. Guxens et al. reported no association between multiple measures of prenatal air pollution and WBV in a large prospective cohort study (6-10 years, Guxens et al., 2018). HV was also mentioned as an outcome, but the results of this analysis were not reported. Van den Dries and colleagues investigated the association between prenatal exposure to organophosphate pesticides and offspring WBV and HV in the same cohort as Guxens et al. No statistically significant associations were reported (10 years, van den Dries et al., 2020). In the cross-sectional study performed by Janulewicz et al., no significant association was reported between prenatal tetrachloroethylene (PCE)-contaminated drinking water exposure and offspring HV, although HV was 1.7% larger in exposed individuals (30 years, Janulewicz et al., 2013).

Disease, treatment and physiology
Ten studies investigated the association between a prenatal maternal disease, treatment and physiology exposure and WBV (N = 5), HV (N = 8) and/or TLV (N = 1) ( Table 1, Supplementary Table 1 and 2). Six additional studies on maternal infections and seven studies on maternal mental health problems are discussed in the subcategories below.
In their twin study on placental functioning, Luo and colleagues reported a negative association between WBV and placental-oxygen time to plateau (TTP) as a measure for placental oxygen transport (GA 29-34 weeks, Luo et al., 2017). Sammallahti et al. reported a negative association between WBV at age 10 years and the umbilical artery pulsatility index during the second trimester as a measure of placental vascular resistance, although this association did not survive correction for multiple testing (Sammallahti et al., 2020). No statistically significant associations with WBV were reported for maternal cancer and chemotherapy (GA 42 weeks, Passera et al., 2019) and maternal hypothyroxinemia (8 years, Ghassabian et al., 2014). Of note, the exposed infants in the study by Passera and colleagues did have a 6.8% smaller WBV. Korevaar et al. studied both increased and decreased maternal thyroid function in the same population as Ghassabian and colleagues. They did not observe a significant association between free thyroxine (FT4) and WBV, but did report a significant inverse U-shaped association between thyroid stimulating hormone (TSH) and WBV (8 years, Korevaar et al., 2016).
A significantly smaller HV was reported after prenatal exposure to maternal iron deficiency anemia at age 3-5 days (Basu et al., 2018). In addition, a significantly larger HV was reported after prenatal exposure to maternal-fetal Rhesus and ABO blood incompatibility at age 40 (Freedman et al., 2011). Willoughby et al. reported a significant negative association between maternal TSH levels in the second and third trimester and right HV (10 years, Willoughby et al., 2014). The five remaining studies investigating prenatal exposure to gestational diabetes (10 years, Jabes et al., 2015), maternal cancer and chemotherapy (GA 42 weeks, Passera et al., 2019), maternal pre-eclampsia (10 years, Ratsep et al., 2016), maternal hypothyroxinemia (8 years, Ghassabian et al., 2014) and FT4 or TSH (8 years, Korevaar et al., 2016) did not report a statistically significant association with HV. Of note, Ratsep et al. did report a larger HV in exposed children (left HV +9.5%, right HV +4.5%) and Ghassabian a smaller HV in exposed children (− 1.9%). However, the statistical analyses were only performed with correction for ICV (Ratsep et al., 2016) or WBV (Ghassabian et al., 2014). Individuals prenatally exposed to pre-eclampsia had a higher TLV compared to unexposed individuals at age 10 (Ratsep et al., 2016).

Infections
We included six studies investigating an association between prenatal maternal infections and HV (N = 5) or TLV (N = 1) ( Table 1,  Supplementary Table 1 and 2). None of these studies included WBV as an outcome.
Ellman and colleagues found no significant association between HV and maternal serum cytokine interleukin-8 levels, a measure of maternal infection (40 years, Ellman et al., 2010). Of note, this analysis had an unusually small sample size of N = 8 and may have been underpowered. Vertical HIV infection had inconsistent effects on HV. Both Nwosu and Yadav and et al. reported smaller volumes of both left and right HV in HIV-infected children at age 7 and 10 (Nwosu et al., 2018;Yadav et al., 2017). However, this effect did not remain statistically significant after correcting for multiple testing in the analysis performed by Nwosu and colleagues. Paul et al. and Wade at al. have an overlap in their study participants, although the extent of this overlap is unclear. Both studies reported no significant association between HIV infection and HV at age 11, although the exposed children had 1% larger HV in the study by Paul and colleagues. In addition, Wade et al. reported no longitudinal changes (follow-up median 53 weeks) in HV associated with HIV infection (Paul et al., 2018;Wade et al., 2019).
Hoffmann and colleagues observed a significantly smaller TLV in fetuses exposed to a cytomegalovirus infection compared to unexposed fetuses, which was independent of WBV (GA 33 weeks, Hoffmann et al., 2010).

Mental health problems
Seven included studies investigated any measure of maternal mental health during pregnancy in association with offspring HV. One of these studies additionally included WBV as an outcome, none included TLV (Table 1, Supplementary Table 1 and 2). Wu et al. reported no significant association between maternal stress, anxiety or depression and WBV (GA 28-36 weeks, Wu et al., 2020b). The children exposed to maternal psychopathology in the study by Björnebekk et al. had a 1.6% larger HV compared to unexposed children, but this difference was not statistically significant (4.5 years, Björnebekk et al., 2015). In addition, no significant associations between maternal depressive symptoms (26 days, Lehtola et al., 2020;GA 28-36 weeks, Wu et al., 2020b;10 years, Zou et al., 2019), maternal stress (23 years, Mareckova et al., 2018;Wu et al., 2020b), maternal anxiety ; 40-66 weeks post-conception, Qiu et al., 2013;Wu et al., 2020b) or maternal cortisol (7.5 years, Buss et al., 2012) and HV were reported. Wu et al. tested for associations between maternal depression, stress, state anxiety and trait anxiety and HV. Of these exposures, only maternal trait anxiety had a significant negative association with HV which remained after correction for multiple testing.
Qiu and colleagues did not identify an association between prenatal anxiety exposure and bilateral HV at birth or six months. They did report slower bilateral hippocampal growth between these time points after exposure to prenatal maternal anxiety. This effect was independent of postnatal maternal anxiety for the right hippocampus only.

Discussion
The 79 studies included in this systematic review provide evidence that prenatal exposures are associated with brain size, especially with WBV and HV. The evidence for an association with TLV is scarce (N = 5 studies). The most substantial evidence was found for smaller WBV and/ or HV after prenatal alcohol exposure, smoking, opioid or cocaine use, nutrient shortage and disease or placental/umbilical cord dysfunction. The outcomes were mostly assessed in childhood and effect sizes ranged between − 40% and + 19%, although many studies failed to report a measure of effect size. Limited evidence was found for other types of prenatal illicit drug exposure, mental health problems, dietary supplementation or physiological ranges, thyroid function, infections or environmental exposures being associated with the outcomes of interest, although evidence was scarce or inconclusive for numerous exposures. Overall risk of bias was low, although studies scored poorly on the description of response rate and characteristics of non-respondents, follow-up adequacy and sample size justification.

Alcohol
The overall body of evidence for a smaller WBV and HV after prenatal alcohol exposure is convincing. Given the large number of studies investigating this exposure (N = 32), the overlap in study sample for some studies will not have significantly impacted the overall outcome. Notably, most studies have measured the outcome between 11 and 14 years of age, which leaves a large knowledge gap in brain development before and after this age range.

Illicit drugs
The included studies provide evidence for a smaller WBV after prenatal opioid and cocaine exposure in neonates, children and teenagers. No studies investigated these associations in (late) adulthood. The smaller group difference in the study by Roussotte at age 15 versus Rivkin at age 12 (-1.4% versus -6.9%) could potentially indicate a dilution of the effect of the prenatal exposure on neurodevelopment, although additional studies investigating these effects in late childhood and (late) adulthood are needed to determine these associations. No associations between prenatal exposure to cannabis, marijuana or methamphetamine and WBV were reported, but the evidence was limited. Evidence was provided for a smaller HV after prenatal opioid exposure. Associations between heroin, cocaine and methamphetamine and HV were heterogeneous and reported by a small number of studies.

Smoking
The included studies demonstrate that the negative association between maternal smoking and offspring WBV is present in the fetal brain and appears to continue into childhood. The evidence for an association between maternal smoking and HV was insufficient. No studies were performed in early and mid-adulthood and the single study that measured HV in late adulthood did not report on WBV, and only reported HV analyses corrected for ICV. It thus remains unclear whether an effect may be observed for WBV or ICV-dependent HV in (late) adulthood.

Diet
Although the included studies on the association between maternal prenatal diet and brain development investigated a large variety of nutrients, they report an overall pattern of smaller WBV after nutrient shortage. This pattern appears to be a general effect across the WBV, few specific effects were observed in the HV. In addition, the effects appear to have sex-specificity, with boys potentially being more vulnerable. The negative correlation of maternal prepregnancy BMI with HV in boys, and the positive correlation of prenatal maternal omega-3 fatty acid intake with HV do point at a potential vulnerability of the hippocampal development to maternal diet.

Environmental
No evidence of associations with prenatal environmental exposures was provided by the included studies. Nevertheless, only three studies were included in this category of which two were conducted in the same cohort, and all administered indirect and inaccurate measures of exposure. More studies investigating the impact of prenatal exposure to harmful environmental circumstances are needed to improve our understanding of the potential association with brain development.

Disease, treatment and physiology
Despite the large diversity in exposures, the included studies provide evidence that maternal medical conditions can impact fetal brain development. In particular, maternal TSH levels, iron deficiency anemia, maternal-fetal rhesus and ABO blood incompatibility and preeclampsia were associated with altered brain outcomes. Surprisingly, larger HVs were observed in individuals exposed to maternal-fetal rhesus and ABO blood incompatibility at age 40 years (Freedman et al., 2011). Furthermore, larger TLVs were observed in individuals exposed to maternal pre-eclampsia (Ratsep et al., 2016). The authors suggest that this may be indicative of adaptive resilience in exposed individuals (Freedman et al., 2011).
4.1.6.1. Infections. The included studies provide some indication that prenatal infections may impact the development of the hippocampus and temporal lobe regions, although evidence is scarce. Most evidence for an association with smaller HV was found in the studies investigating vertically transferred HIV, although study results were inconsistent. 4.1.6.2. Mental health problems. No evidence for an association between mental health problems and WBV or HV was provided by the included studies. Nevertheless, most studies on HV only report their outcomes corrected for either ICV or WBV, so we cannot infer that HV developmental alterations are dependent on ICV or WBV. The large majority of studies was performed in infants and no studies assessed the outcome past 10 years. This limits the available information on developmental brain trajectories after prenatal exposure to maternal mental health problems. The reduced right hippocampal growth independent of postnatal mental health problems reported by Qiu and colleagues does indicate that some association may be present. Additional studies are needed to elucidate this potential association, especially since this study was performed in young children only.

General interpretation
Taken together, these studies provide a convincing body of evidence for an association between adverse prenatal exposures and smaller brain volumes in brain regions associated with AD. However, several nuances should be considered. First, the specific impact of any exposure depends on the type of exposure and the underlying mechanisms which drive the effect. Several prenatal exposures had a clear impact on brain volumes, whereas others showed no hint of an association. Furthermore, the effect size is likely dependent on the severity and timing of the exposure. Little attention is given to the timing of exposure in the majority of included studies. Moreover, many exposures may last throughout pregnancy and continue postnatally, providing challenges in determining the effect at specific points in pregnancy. For instance, by definition, populations of children vertically infected with HIV are continuously exposed postnatally. In addition, there may be sex-specific effects for certain exposures, as was demonstrated by De Rooij et al. and Alves et al. (Alves et al., 2020;de Rooij et al., 2016). Nevertheless, outside of the diet category, included studies scarcely explored sex-specificity of their results. Exploring sex-specificity of the impact of prenatal exposures on brain development may provide valuable insights, especially considering that male fetuses are more vulnerable to prenatal exposures and may therefore respond differently (Bale, 2016).
Furthermore, numerous study groups included multi-exposures, which may be problematic for determining the effect of a single, specific exposure. In particular, prenatal exposure to alcohol, smoking and/ or illicit drugs were often observed in combination. Correcting for multiexposures remains challenging, especially when exposure information is based on self-report. Toxic effects of multiple exposures may amplify their impact, resulting in larger effect sizes. Of note, exposure to multiple, interrelated exposures is common in real life, and studying their combined impact may improve the external validity of the study (Buss and Genueit, 2022). Thereby, both studies of single exposures and studies investigating multi-exposures should be considered to obtain a more complete understanding of the impact of prenatal exposures.
Most studies measured the outcome in childhood, leaving a large knowledge gap of the outcomes in (late) adulthood. Despite the limited number of studies that measured the outcome in adulthood, the majority of brain development occurs in early development, and brain volumetric measures remain relatively stable thereafter until the onset of neurodegeneration (Dobbing and Sands, 1973;Prayer et al., 2006). Large changes in effect size in later life are, therefore, not expected.
The effect size (Δexposed-unexposed, %) in the included studies ranged between − 40% and + 19%, although many studies failed to report any indication of effect size or brain volumes. The approximated average effect of alcohol (~− 10%) on brain volumes was roughly twice as large as the average effect of illicit drugs (~− 6%) or smoking (~− 5%) in the studies included in this review. These effect sizes are likely to be relevant alterations to the brain reserve, as studies generally report a 10-15% smaller HV in MCI patients and a 15-40% smaller HV in AD patients compared to healthy controls (Bosscher and Scheltens, 2002;Pini et al., 2016). Thereby, having a 5-10% smaller brain volume may significantly impact the brain's capacity to buffer AD-related neurodegeneration.

Mechanisms
Several biological pathways through which adverse prenatal exposures could result in altered brain development have been proposed. These pathways are mainly associated with teratogenic effects on the developing fetus or nutrient shortage limiting developmental processes (Martin-Gronert and Ozanne, 2012).
For instance, a prenatal exposure or nutrient shortage may alter DNA methylation patterns of genes associated with prenatal brain development and growth in general (Tobi et al., 2014). In addition, prenatal malnutrition has been shown to affect neurogenesis, cell migration and differentiation (Morgane et al., 1993). Comparable processes have been observed for prenatal alcohol toxicity disrupting neuronal proliferation and migration, causing cell death. Furthermore, alcohol-induced hypoxia and altered hormone and protein synthesis levels can result in growth retardation, and alcohol can disrupt growth factor signaling and increase oxidative stress on the embryo . Similarly, prenatal tobacco exposure can induce neural cell loss and hypertrophy (Scott-Goodwin et al., 2016). Rodent studies have highlighted the effects of prenatal exposures on brain reserve through alterations in dendritic morphology, spine number, and synaptic plasticity and function (Lesuis et al., 2018). Moreover, a primate study of fetal nutrient restriction related resulting cerebral developmental disturbances to mechanisms including impaired cell proliferation, glial maturation and neuronal process formation (Antonow-Schlorke et al., 2011).

Methodological remarks
After thorough evaluation of the included studies, several methodological aspects deserve mentioning.
First, many studies analyzed HV corrected for WBV or ICV, without reporting an uncorrected analysis or any information on WBV in the study participants. Consequently, these studies provide little information on the impact of the exposure of interest on brain development as a whole, as a reduction in HV dependent on WBV would still be of interest. Furthermore, additional reporting on WBV would greatly improve the insight into brain development in exposed individuals.
Reporting on study outcomes was incomplete or unclear for many studies. Numerous studies failed to report brain volumes, uncorrected analyses and effect sizes. As a result, we were limited to a mere report of statistical significance if no additional information was provided. As statistical significance does not give any insight into effect size or potential clinical relevance, we extracted outcome details if provided and used this as context for all studies in the exposure category. In addition, we calculated a % difference between exposed and unexposed groups, when possible, to enable some interpretation of effect size. Furthermore, terminology for WBV was confusing. Reporting of WBV, total brain volume, intracranial volume with or without ventricles or cerebellar volume was often inaccurate and could easily be misunderstood. Studies regularly failed to report whether HV was reported or analyzed unilaterally, bilaterally or as an average of both hemispheres. Since studies have previously identified specific effects for either left HV or right HV, clear reporting of the analysis approach and rationale is of utmost importance.
Additionally, there was considerable overlap in the study samples of numerous studies. Several large cohort studies were used for the analysis of multiple exposures or different outcomes on the same exposure. Also, some manuscripts combined samples from previous studies to explore a new research question in a larger study sample. In the case of considerable overlap in study sample, exposure and outcome, the study with the lowest number of participants was excluded. However, if, despite an overlap in study participants, new information was provided by the study in terms of outcome or age at outcome assessment, studies were not excluded. We aimed to clearly report on these details in both text and tables to enable an unbiased interpretation of the study results.
Moreover, many included studies obtained information on exposure through maternal self-report or retrospective reports sometimes collected years after the prenatal period. This approach reduces the reliability of the exposure data and increases the chances of potential underreporting and misclassification of exposure. Nevertheless, although increasing the risk of bias in a study, underreporting and misclassification would presumably lead to an underestimation of effects. Therefore, the overall pattern of reduced brain volumes after harmful exposures reported supports the hypothesis of an association despite this limitation.
In addition, some exposure categories only included a small number of studies, and the evidence for many specific exposures was limited to a single study. Overall, the number of studies per specific exposure type should be increased to improve the certainty of evidence, especially in categories with single studies per exposure.
Lastly, within exposure categories, we identified considerable differences between studies in exposure definition and severity limiting comparability between studies. Also, most studies in the alcohol and illicit drug exposure categories used a relatively severe definition of exposure and occasionally included mildly exposed individuals in the unexposed control group. Additionally, many prenatal alcohol exposure studies investigated a specific population of children with FAS, a syndrome defined by significant brain damage. Combined, these factors limit the external validity to societies where pregnant women generally restrict their substance use to for instance an occasional glass of wine. Future studies should shed light on the impact and effect size of sporadic substance use versus complete abstinence on offspring brain development.

Strengths and limitations
In this systematic review, we provide an overview of the existing body of literature on prenatal exposures and outcomes of WBV, HV and TLV. By selecting studies based on these outcomes instead of AD diagnosis, we were able to identify a large number of studies with outcomes highly relevant for AD, which have not been related to risk of developing AD. This innovative approach resulted in valuable insights for future research strategies, and substantiates previous suggestions of a potential contribution of the prenatal environment to the risk of developing AD. Nevertheless, only a limited number of studies was identified investigating the outcomes in older age (N = 3 above 50 years), and more longitudinal studies are needed to map the impact of prenatal exposures on brain health in later life.
Several limitations are the result of the nature of the exposures and outcomes of interest and our study design. First, several of the exposures discussed above are not specific to the prenatal period. For instance, maternal mental health problems or environmental exposures may continue postnatally, and disentangling these effects is challenging. Secondly, the outcomes included in this systematic review are established MRI neuroimaging biomarkers for AD. They may, however, lack specificity for AD, as they are also associated with other types of dementia and neurodegeneration. For instance, HV atrophy is also observed in non-AD forms of dementia including Parkinson, vascular, frontotemporal lobar and semantic dementia (Pini et al., 2016;Whalley et al., 2006). As a result, despite having limited specificity for AD, the conclusions of this systematic review may also be applicable to other types of dementia and late-life neurodegenerative disease. Lastly, we were unable to perform a meaningful meta-analysis because of the large variation in exposures, outcomes and age at outcome assessment in the included studies. These limitations also restricted the comparability between studies. Nevertheless, grouping the studies per exposure category facilitated comparison within categories. The large range in exposures and outcomes did result in a substantial body of literature with a broad external validity, enabling general conclusions for a broad population.
The nature of the research topic and the observational designs of included studies restricts conclusions on causality. It is nearly impossible to rule out all potential confounders and randomized controlled trials are scarce. One included study did adopt a randomized controlled trial design, and reported a higher WBV in men after essential brain specific fatty acid supplementation, providing evidence for a causal relationship (Ogundipe et al., 2018).

Recommendations for future research
This systematic review demonstrated a clear association between prenatal exposures and brain development. Our conclusions on the progression of these outcomes over time and late-life brain health are, however, speculative. As the number of studies that longitudinally assessed the outcomes of interest was limited, and only a small number of studies assessed the outcomes past childhood. Cross-sectional studies provide a limited view of brain development, since measuring at a single time point does not provide any information on potential growth retardations and catch-up effects. There is a need for longitudinal studies stretching across a broader age range to explore the impact of prenatal exposures throughout life. Furthermore, core outcome sets and reporting guidelines could greatly improve the usability of future studies. Ideally, studies should include both WBV and HV as outcomes, and report on volumes, effect size and uncorrected models in addition to corrected models.
Finally, we urge investigators in future studies to adopt a life-long perspective in studying AD. The evidence summarized in this systematic review endorses the hypothesis that the prenatal environment may be an essential factor in the development of brain regions associated with AD risk in late life. Embedding this realization in research into both prenatal development and brain aging may promote collaborations among researchers in both fields and facilitate breakthroughs which can significantly move the field forward.

Conclusions
Adverse prenatal exposures are associated with alterations in brain development measured in structural brain outcomes related to AD in late life. Specifically, prenatal exposure to alcohol, opioids, cocaine, nutrient shortage, placental dysfunction and maternal anemia were associated with smaller whole brain, temporal lobe and hippocampal volumes. This altered development may result in decreased brain reserve which is associated with an increased risk of dementia. Despite the relatively high overall quality of the included studies, there was a considerable range of exposures, outcomes, study designs and ages at outcome assessment, and reporting was inconsistent. Following existing neuroimaging reporting guidelines such as the OHBM COBIDAS Report is highly recommended (https://www.humanbrainmapping.org/files/2016/COBI-DASreport.pdf; Pernet et al., 2020). In addition, core outcome sets and reporting guidelines for this field could improve the comparability between studies, overall research quality and applicability of research in this field. Thereby, the field of neuroimaging after prenatal exposures may offer a true life course perspective on the development of AD and other neurodegenerative diseases in later life. Children from a peri-urban area in South Africa (no prenatal illicit drug exposure, birth >35 weeks GA, Apgar score >6 at 5 min, no encephalopathy or neonatal complications, no genetic syndrome or congenital abnormality).

Alcohol
Infants with no significant history or biological evidence of substance exposure.
All questionnaires on the DCHS were administered by an interviewer (study nurse or research assistant trained in the questionnaire and in interviewing technique) because of low levels of literacy in the population as a whole. Alcohol exposure was defined as a minimum score of 11 (moderate to high risk of experiencing severe problems as a result of their current pattern of use) on the alcohol questions on the ASSIST questionnaire and a positive history of alcohol use in any trimester of pregnancy at levels consistent with WHO moderate-severe alcohol use (either drinking 2 or more times a week or 2 or more drinks per occasion).  In Los Angeles, alcohol-exposed subjects were recruited primarily through the UCLA Fetal Alcohol and Related Disorders Clinic. In Cape Town, alcoholexposed subjects had participated in a previous study on the epidemiology of FASD among first graders and were recruited through the University of Cape Town from a small city and its surrounding rural areas in the Western Cape Province. In San Diego, alcoholexposed participants were recruited through the SDSU Center for Behavioral Teratology.
Alcohol Control subjects recruited from ongoing studies at each site or specifically for this study via advertisements, word of mouth, or national registers.   Children recruited from three different study sites through university hospital clinics, from a diverse ethnic background. Exposed children were classified into FAS or confirmed exposure without physical features of FAS.

Alcohol
Children screened for prenatal alcohol exposure and not exposed to more than 1 drink per week on average or more than two drinks on any one occasion during pregnancy.  Exposed individuals referred from local FASD Clinics or from ads with local FASD-specific caregiver groups, or recruited through online and print fliers, community forums, mouth form former participants.

Alcohol
Unexposed individuals with no or minimal exposure. The same exclusion criteria were used for the exposed and control groups.
Review of records or maternal report and confirmation by a licensed medical doctor. Alcohol exposure was defined as an average of > 13 drinks per week or > 4 alcoholic drinks per occasion at least once a week. Minimal alcohol exposure was defined as no more than 1 drink per week on average and never more than 2 drinks per occasion. FASD was diagnosed by experienced multidisciplinary teams using the Canadian Guidelines. Alcohol history was confirmed by reliable sources, birth records, child welfare report, legal documents and direct interviews with the birth mother.

NeuroDevNET FASD project
Tables 1a-h are very large and make the text quite unreadable if they're placed in-text like this. Could they be placed as a single table (as submitted) after the conclusions? Also, please place them in landscape view as submitted. We believe this will significantly improve readability.  Hawaii, USA Nonalcohol-exposed, healthy children (singleton pregnancy, maternal age ≥18 y, English speaking, no single illicit drug use besides methamphetamine or infection during pregnancy.

Methamphetamine (tobacco)
Unexposed infant-mother pairs who denied methamphetamine use and had a negative meconium screen matched for maternal race, birth weight, private vs public insurance, education.
Standardized maternal self-report at birth and 1 month and/or a positive meconium sample screened with gas chromatography/mass spectroscopy.

Cannabis (tobacco)
Unexposed controls matched for gender and a fuzzy match for age (4-month range).
Maternal self-report with a questionnaire during the first trimester (use before or during pregnancy, still using, product used and frequency) and maternal THC levels from urine samples in early, mid and late pregnancy (subset of the cohort). Opioids (other illicit drugs, poly-drug) Unexposed children with minimal biomedical vulnerability and social factors recruited from a nonclinical setting of local maternal and child health centers (no unusual sleep or feeding problems during the first 6 months, maternal alcohol, illicit drug or tobacco (daily) use).
Referral because of concerns about illicit drug abuse, maternal medical and social records and self-reports.

Perinatal risk project
Riggins 2012 138 69 USA Exposed infants recruited during their postnatal stay in a university hospital that served a largely inner-city, African American population (GA >32 weeks, birth weight >1750 g, no congenital or medical problems requiring admission to the neonatal intensive care unit).
Heroin and/or cocaine Two unexposed community comparison samples born in the university hospital during the same period as the exposed children, recruited from the university primary care clinic at the 5-year time point and early adolescence time point by medical record review, matched for SES, age of first pregnancy, and race.
Positive maternal toxicology screen, positive infant toxicology screen, maternal self-report, and/or notation in the mother's chart at delivery. Heavy exposure was defined as a positive toxicology screen at birth and/or maternal self-reported use of 2 times or more per week during the last 6 months of pregnancy.

NA
(continued on next page) A. Boots et al. Hospital based population referred to pediatric department between 1997 and 2012 (no FASD, opioid maintenance treatment).

Opioids
The first child of the same gender born at the same hospital on the same date, with a birth weight above 3000 g was invited to participate. If they declined, the next child was contacted.
Infant admission to the neonatal department due to maternal illicit drug use, in most cases treated for withdrawal symptoms, or referral to a pediatric neurologist at a later age with a medical history of prenatal illicit drug exposure and symptoms of attention and/or behavioral problems. Mothers of controls filled in questionnaires.

Maternal hypothyroidism
Typically developing children born to mothers without hypothyroidism in the same period as the exposed group (full-term birth, no learning disability or attention disorder) recruited from a local hospital nursery, the Motherisk database or a local obstetrician.
Exposed: TSK and FT4 measurements. Controls: Interview and questionnaires.

Maternal iron deficiency anemia
Unexposed healthy, inborn, term singleton neonates. Matched to exposed group for GA.
Blood was collected after complete delivery of the neonate. Severity of maternal anemia was graded as mild  HIV-infected children identified from programs for the prevention of mother-tochild transmission of HIV-1 in the Western Cape and Gauteng provinces. Infected mothers were also treated for HIV infection.

(Maternal) HIV infection
Uninfected controls recruited from a vaccine trial. These included both children born to HIV+ mothers (HIV-exposed but uninfected children) and to uninfected mothers (HIV-unexposed and uninfected children HIV infected children with a CD4% of 15-24%, (no history of AIDS illness, anti-retroviral therapy except for exposure as part of prevention of mother-to-child transmission, active AIDS-defining illnesses or previous use of immunosuppressive drugs, immunomodulators within 30 days of study entry, abnormal laboratory results, prior or current brain infection, neurological or psychiatric disorder, congenital abnormality or head injury with a loss of consciousness).

(Maternal) HIV infection
Uninfected controls, including healthy unexposed uninfected (HUU) and HIV-exposed but uninfected children (HEU). HIV infected children with a CD4% of 15-24%, (no history of AIDS illness, antiretroviral therapy except for exposure as part of prevention of mother-to-child transmission, active AIDS-defining illnesses or previous use of immunosuppressive drugs, immunomodulators within 30 days of study entry, abnormal laboratory results, prior or current brain infection, neurological or psychiatric disorder, congenital abnormality or head injury with a loss of consciousness).
(Maternal) HIV infection HIV-exposed but uninfected children and HIV unexposed children combined into one control group as researchers did not find significant differences in brain imaging indices in previous studies. HIV assays were performed as per the national HIV testing protocols: screening by HIV enzyme-linkedimmunosorbent-assay (ELISA)/Rapid test followed by confirmation with 2 further HIV rapid tests of higher specificity.
NA a Included in analysis of interest. b Included in analysis of interest. Exposures between brackets are additional exposures included in the study as multi-exposures. c Only controls are of interest. Specific patient group data not included.

Declaration of Competing Interest
none.