Early study of the genetic determination of BW

BW is commonly documented as an indicator of perinatal health and has been associated with risk of adult cardiometabolic disease through epidemiological analyses. ^{Reference Alambert, de Gusmão Correia, Rajendram, Preedy and Patel1,Reference Hoffman, Reynolds and Hardy2} It is a complex trait influenced by multiple factors. Both maternal and paternal BW represent significant predictors of offspring BW, ^{Reference Hackman, Emanuel, Van Belle and Daling18–Reference Magnus20} and associations have also been found for multiple additional maternal metabolic parameters and environmental factors. ^{Reference Breschi, Seghieri, Bartolomei, Gironi, Baldi and Ferrannini21–Reference Sacks, Liu, Wolde-Tsadik, Amini, Huston-Presley and Catalano23} The determination of BW is also contingent on a significant genetic component, which is estimated to explain around 25%–60% of its variance, ^{Reference Magnus20,Reference Magnus24–Reference Mook-Kanamori, van Beijsterveldt and Steegers27} (single nucleotide polymorphism-heritability estimated around 11%–30%). ^{Reference Horikoshi, Beaumont and Day28–Reference Warrington, Beaumont and Horikoshi30} The link between genes influencing both BW and health outcomes was first elucidated through the investigation of individual genetic variants influencing growth and metabolism.

Having established an association between low BW and type-2 diabetes (T2D) risk, ^{Reference Hales, Barker and Clark31,Reference Phipps, Barker, Hales, Fall, Osmond and Clark32} it was hypothesized that impaired function of genes regulating glucose metabolism might influence intrauterine growth through the dual action of insulin as growth factor and regulator of glucose metabolism. Hattersley et al. ^{Reference Hattersley, Beards, Ballantyne, Appleton, Harvey and Ellard33} demonstrated that both parental and fetal genotype altered fetal growth and postnatal metabolism independently by studying monogenic mutations in the glucokinase (GCK) gene. Such mutations are a cause of Maturity Onset Diabetes of the Young (MODY), a monogenic form of diabetes. In a sample of 58 offspring, inheriting a defective copy of the GCK gene resulted in a mean BW reduction by 533 g, compared to offspring with two functional copies. In contrast, maternal hyperglycaemia caused by the same mutation, but not inherited, increased the offspring’s mean BW by 601 g. These effects were additive; in the cases in which both mother and offspring carried a defective GCK gene, the mean offspring BW was not significantly different from that of mother–offspring pairs without the mutation. In addition, in a sample of 21 sibling pairs discordant for the presence of a defective GCK copy, the children with the mutation had a 521 g lower mean BW. The later investigation of a polymorphism in the GCK promoter region that had been associated with T2D risk, identified that the maternal allele was associated with a mean increase of 64 g in offspring BW. ^{Reference Hattersley, Beards, Ballantyne, Appleton, Harvey and Ellard33} However, there was no apparent effect on BW associated with the fetal genotype at this polymorphism. ^{Reference Weedon, Frayling and Shields34,Reference Pawlowski, Kouklinos and Jopp-Petzina35}

These findings introduced the “fetal insulin hypothesis,” ^{Reference Hattersley and Tooke36} describing a mechanism of BW determination alternative to the DOHAD concept. According to Hattersley and Tooke, the association between low BW and adult disease is, mainly, genetically mediated by diabetes risk alleles. Fetal variants impairing insulin production or increasing insulin resistance would not only cause reductions in fetal growth, but also cascade into abnormal development and later disease. This study also pointed out the independent effect of fetal genetics (influencing glucose sensing and uptake) and maternal genetics (influencing intrauterine glucose homeostasis) on BW, often difficult to discern due to genotype correlation and confounding. These results shifted the focus of the developmental programming community into investigating the effect of T2D-associated alleles on BW.

Table 1.

Chronological summary of studies investigating the genetics behind birth weight

BW, birth weight; GA, gestational age; M/O, mother/offspring; T2D, type 2 diabetes; CAD, coronary artery disease; FPG, fasting plasma glucose; SBP, systolic blood pressure; SNP, single nucleotide polymorphism; VNTR, variable number tandem repeat; GWA, genome-wide association; EWA, epigenome-wide association; ANOVA, analysis of variance; GCK, glucokinase; IGF2, insulin-like-growth-factor-II; TCF7L2, transcription factor 7-like 2; ADCY5, adenylyl cyclase 5.

Variants in glucose metabolism genes and association with BW

Further studies compared the effect of polymorphisms in glucose metabolism-relevant genes on early growth, with variable results. Results regarding the association with BW of a polymorphism in the insulin gene were conflicting, ^{Reference Dunger, Ong and Huxtable37,Reference Bennett, Sovio and Ruokonen38} and no direct statistical association between BW and other candidate genes (IRS-1, HNF-1α, NHNF-4 α, HNF-6, KCNJ11, PPARγ2) was found. ^{Reference Rasmussen39–Reference Bennett, Sovio and Ruokonen42} However, association with BW was found for polymorphisms in the insulin-like-growth-factor-II (IGF2) and its receptor IGF2R, ^{Reference Kaku, Osada, Seki and Sekiya43} and polymorphisms in the IGF2 regulator region H19 ^{Reference Petry, Ong and Barratt44} and the transcription factor 7-like 2 (TCF7L2) gene, an important variant associated with T2D susceptibility. ^{Reference Zhang, Qi and Hunter45} Using a sample of 8344 mothers and 15,709 offspring, Freathy et al. ^{Reference Freathy, Weedon and Bennett46} determined a mean increase of 30 g in offspring BW for each maternal copy of the TCF7L2 T2D-predisposing allele, and a mean increase of 18 g in BW for each fetal copy. Analysis of 6044 mother–offspring pairs showed that this association was driven by the maternal genome, the increase in offspring BW most likely a product of limited maternal insulin secretion, resulting in maternal hyperglycaemia. The combined maternal genotypic effects of weight increasing alleles in both the TCF7L2 and the GCK genes were estimated to have similar impact on offspring BW as smoking four cigarettes per day during the third trimester of pregnancy, ^{Reference Bernstein, Mongeon, Badger, Solomon, Heil and Higgins47} albeit directionally opposed.

With the determination of multiple loci involved in T2D pathogenesis and risk-prediction through genome-wide association, ^{Reference Prokopenko, McCarthy and Lindgren48} a new set of candidate variants with possible association with fetal growth were pursued. Pulizzi et al. ^{Reference Pulizzi, Lyssenko and Jonsson49} identified correlations between BW and common T2D risk variants HHEX and CDKN2A/B. However, the lack of maternal genotype data made it impossible to determine if these represented actual fetal contributions or confounding from correlation with the maternal genotype. Freathy et al. ^{Reference Freathy, Bennett and Ring50} replicated these results using data from 7986 mothers and 19,200 offspring from four different cohorts, testing five loci (CDKAL1, HHEX-IDE, CDKN2A/B, IGF2BP2, and SLC30A8) with T2D-risk increasing variants for association with BW. From these, only fetal CDKAL1 and HHEX-IDE loci showed significant association, with a combined 17 g reduction in BW per additional copy of the risk allele. This association was reported to be driven almost exclusively by the fetal genotype, because the influence of the maternal genotype on offspring BW (regarding the studied polymorphisms at CDKAL1 and HHEX-IDE) appeared to be minimal and directionally opposed. Zhao et al. ^{Reference Zhao, Li and Bradfield51} performed a similar investigation, exploring the association of additional single nucleotide polymorphisms (SNPs) at 20 T2D-susceptibility related fetal loci, including CDKAL1, HHEX-IDE, CDKN2A/B, and TCF7L2, with BW. Using data from a pediatric cohort of 5465 children, only a single variant at CDKAL1 was found to be significantly associated with BW. These results support those of Freathy et al. ^{Reference Freathy, Weedon and Bennett46,Reference Freathy, Bennett and Ring50} regarding the significant contribution of fetal CDKAL1 to BW, and the not statistically significant contribution of fetal variants at CDKN2A/B, IGF2BP2, and TCF7L2. Adkins et al. ^{Reference Adkins, Somes and Morrison52} re-evaluated the contribution of IGF2, IGF2R, and H19 on fetal growth using three independent American populations, finding mixed agreement with previous studies of IGF2 and IGF2R concerning the presence of an association with BW depending on the source of the effect (fetal or maternal), the SNP in question, and the sample size and ancestry. ^{Reference Kaku, Osada, Seki and Sekiya43,Reference Gomes, Soares, Pasqualim-Neto, Marcondes, Lôbo and Ramos53,Reference Heude, Ong, Luben, Wareham and Sandhu54}

Independent maternal and fetal genetic effects: correlation and confounding

The study of glucose metabolism-related genes and their influence on BW identified distinct and independent effects for fetal and maternal variants, as evidenced by some of the studies presented above. ^{Reference Hattersley, Beards, Ballantyne, Appleton, Harvey and Ellard33,Reference Weedon, Gloyn, Davey Smith, Ben-Shlomo, Frayling and Hattersley40,Reference Kaku, Osada, Seki and Sekiya43,Reference Petry, Ong and Barratt44,Reference Freathy, Weedon and Bennett46,Reference Freathy, Bennett and Ring50} Maternal T2D-related genetic variants often influence offspring BW with an effect directionally opposed to that of the same fetal variants. The interplay between intrauterine glucose concentrations, mediated by the maternal genotype and environment, and possible insulin impairments inherited by the offspring, results in the observed variability in BW as reflected in the fetal insulin hypothesis. ^{Reference Hattersley and Tooke36} However, despite the clear duality of effects, the actual estimation of their magnitude on BW is complex, because the maternal genotype might influence offspring BW through her own metabolic phenotype and through direct inheritance by the fetus. The correlation between maternal and offspring genotypes only adds to this complexity. The fetal genotype is estimated to explain around twice the variance in BW explained by the maternal genotype, ^{Reference Magnus24,Reference Lunde, Melve, Gjessing, Skjaerven and Irgens26,Reference Beaumont, Warrington and Cavadino29,Reference Magnus55–Reference Hughes, Nodzenski and Beaumont57} figures greatly depending on the polymorphism investigated and the population sample chosen. The current increase in genome data availability from mother–offspring pairs and the prevalence of methodologies such as genome-wide association (GWA) analyses, ^{Reference Wang, Lu and Zhao16} structural equation modelling, ^{Reference Jak58,Reference Warrington, Freathy, Neale and Evans59} and Mendelian randomization (MR), ^{Reference Davey Smith and Ebrahim60,Reference Pierce and Burgess61} holds promise of not just expanding our understanding of genetic variants influencing fetal growth, but also unravelling the distinct effects of maternal and fetal genome on BW.

Studies of genome-wide association with BW and adult health outcomes

With the development of consortia such as the Early Growth Genetics (EGG) and the Early Life Genetics and Lifecourse Epidemiology (EAGLE), researchers began taking advantage of GWA analyses combining results from multiple cohorts and studies, to elucidate the genetics underpinning perinatal and childhood phenotype and their association with adult disease. This increase in sample size and scope promoted the identification of numerous BW-related loci in the following decade. Using data from six European cohorts (replicated in 13 studies), Freathy et al. ^{Reference Freathy62} identified SNPs in two fetal loci in chromosome 3 strongly associated with BW. The first polymorphism was located near the LEKR1 and Cyclin-L1 (CCNL1) genes, which had not been formerly associated with fetal growth and whose mechanism of involvement is still unknown. The second polymorphism identified was located in the adenylyl cyclase 5 (ADCY5) locus, a gene related to insulin secretion and adult T2D risk. ^{Reference Dupuis63} When testing the association of previously reported T2D susceptibility and fasting glucose related loci ^{Reference Prokopenko, McCarthy and Lindgren48,Reference Dupuis63} with BW, only ADCY5 and CDKAL1 showed significant association, which might hint at these genes having a more pronounced effect on beta-cells during development.

In a subsequent European-ancestry meta-analysis study, Horikoshi et al. ^{Reference Horikoshi64} identified seven fetal loci presenting association at genome-wide level with BW, including ADCY5, CDKAL1, and one near CCNL1. In the case of ADCY5 and CDKAL1, the BW-lowering allele was correlated with increased T2D risk. Revisiting additional 47 T2D-related loci showed strong correlations with BW in polymorphisms at HHEX-IDE and KCNQ1, lowering BW, and at MNTR1B and GCK, increasing BW as a likely effect of maternal genotype correlation. Similar to CDKAL1 and HHEX-IDE, ^{Reference Freathy, Bennett and Ring50} KCNQ1 and MNTR1B were also related to glucose regulation. ^{Reference Kasuga65,Reference Liao, Liu and Tan66} The other four novel loci identified were ADRB1, HMGA2, LCORL, and a SNP at location 5q11.2. The studied SNP at ADRB1 (rs1801253) had been previously associated with adult blood pressure ^{Reference Johnson, Newton-Cheh and Chasman67} and the studied BW lowering allele was associated with lower blood pressure in adulthood. Studying 29 additional blood pressure-related loci showed no significant association with BW. HMGA2 and LCORL have been associated with adult height and other growth attributes during childhood. ^{Reference Weedon68,Reference Weedon69} After testing another 178 published height-related loci ^{Reference Weedon68} for associations with BW, 132 showed the same direction of effect, although these were not statistically significant. Finally, the polymorphism at 5q11.2 had not been previously linked to fetal growth and, similarly to the CCNL1 locus, its function determining BW remains unknown. The next study by Horikoshi et al. ^{Reference Horikoshi, Beaumont and Day28} featured 153,781 individuals from 37 studies representing multiple ancestries, and identified genome-wide association with BW for 60 fetal loci, from which 53 were novel. Although these 60 distinct fetal signals explained around 2% of the variance in BW, the whole genotyped array captured around 15% of its variance, supporting the theory of BW being determined by a large number of variants with small effects. Many of these signals had been previously associated with cardiometabolic traits, and investigating the biological pathways involved in these associations through gene set enrichment showed significance for glucose and lipid metabolism, growth, and development pathways. These signals showed inverse genetic correlation with disease indicators systolic blood pressure (SBP), coronary artery disease (CAD), and T2D, and strong positive genetic correlation with other anthropomorphic traits such as birth length, adult height, and adult body mass index (BMI). Most BW-raising alleles (such as loci near NT5C2, FES, NRIP1, and PTH1R) were associated with lower adult blood pressure and lower risk of CAD. BW-lowering alleles (such as loci near ADCY5, CDKAL1, and HHEX-IDE) were associated with increased T2D-risk via reduced insulin secretion. Using a sample of 12,909 mother–offspring pairs, it was estimated that that most of these associations were driven by the fetal genotype. For five of the 60 signals (EBF1, L3MBTL3, MNTR1B, FES, and ATCL9), it was estimated that the association with BW reflected an effect from the maternal genotype. However, this method of estimation only compared effect sizes without considering correlation between both genotypes.

To address the distinct influence of the maternal genotype over offspring BW, Beaumont et al. ^{Reference Beaumont, Warrington and Cavadino29} performed structural equation modelling (SEM), which analyses associations for the maternal genotype conditioning on their offspring genotype, and provides increased statistical power to estimate maternal and offspring effects working oppositely. ^{Reference Warrington, Freathy, Neale and Evans59} Using a European sample of 86,577 mothers, 10 maternal loci reaching genome-wide association with offspring BW were identified. SEM analysis using a sample of 8705 mother–offspring pairs singled out maternal effects, acting through the intrauterine environment and independent on the fetal genotype, for seven of these loci (MTNR1B, EBF1, ACTL9, KCNAB1, GCK, TCF7L2, and CYP3A7). These offspring BW-altering polymorphisms identified were located in loci previously associated with glucose metabolism and T2D-susceptibility (MTNR1B, GCK, TCF7L2), hormonal regulation (CYP3A7), and potentially involved in blood pressure (EBF1) and immune function (ACTL9), furtherly supporting the links between early growth and adult disease outcomes. These results reiterate the importance of discerning the maternal contribution to offspring BW.

Following on this avenue of investigation, Warrington et al. ^{Reference Warrington, Beaumont and Horikoshi30} identified 190 association signals (129 of which were novel) in a GWA study meta-analysis of BW, using an European sample of 230,069 mothers and 3,213,223 offspring. The analysis of the fetal genotype identified 146 independent SNPs associated with BW, and the study of the maternal genotype identified 72 SNPs associated with offspring BW, with 30 loci being identified in both analyses, most tagging the same signal. SEM analysis identified 64 SNPs with fetal-only effects, 32 SNPs with maternal-only effects, 27 SNPs with directionally concordant fetal and maternal effects, and 15 SNPs with directionally opposed effects. In addition, the genome-wide genetic correlation between BW and adult traits was calculated, using linkage disequilibrium score regression, independently for fetal and maternal SNPs. Many traits, such as most defining anthropometry (adult and childhood height and BMI), were typically equally correlated with BW for fetal and maternal genotypes. Other traits presented maternal and fetal genetic correlations of different magnitude, for example, SBP having a stronger maternal correlation. Some traits showed genetic correlations working in opposite directions, such as fasting glucose levels having a positive correlation with the maternal genotype, but negative correlation for the fetal genotype. An early maternal genome-wide complex trait analysis (M-GCTA) by Eaves et al. ^{Reference Eaves, Pourcain, Smith, York and Evans56} estimated that nearly 40% of the variance in BW was explained by the combined effects of the fetal (28.5%) and maternal genotypes (7.6%), and their covariance (3.7%). However, due to the genetic correlation between maternal effects on gestational duration ^{Reference Zhang, Feenstra and Bacelis70} and offspring BW, and lacking this information for a large proportion of the individuals made it impossible to rule out that some of these maternal effects on BW operated through alterations in gestational length. However, these findings are very similar to those obtained in a more recent M-GCTA, by Qiao et al. ^{Reference Qiao, Zheng and Helgeland71} performed on over 12,000 mother–offspring pairs from two European cohorts, and adjusting BW for gestational duration. They estimated that fetal genetic variability explained 27% of the variance in BW, with maternal genetics explaining 7% of BW variance, and 5% being explained by the covariance between fetal and maternal genotypes.

MR is a statistical tool that allows inference of causal relationships between an exposure, mediated by a genetic variant as an instrumental variable, and an outcome. One of the challenges of using this tool in the context of DOHaD is identifying a robust genetic instrument for the exposure of interest. Nevertheless, studies performing MR to investigate the causal genetic relationship between BW and multiple cardiometabolic outcomes identified positive causal associations for BMI and atrial fibrillation, and negative associations for T2D, glucose levels, cholesterol, and CAD. ^{Reference Wang, Huang and Li72–Reference Kember, Levin and Cousminer76} However, these conclusions should be interpreted with caution: in addition to obtaining quality birth records with detailed neonatal phenotype, one of the biggest obstacles of MR analysis for birth outcomes is properly accounting for the influence of the maternal genotype on fetal growth. ^{Reference Warrington, Beaumont and Horikoshi30,Reference Lawlor, Richmond and Warrington77–Reference D’Urso, Wang, Hwang, Moen, Warrington and Evans79} The previously presented studies were limited by the lack of maternal genotype data, failing to account for its correlation with the fetal genotype, and obscuring any possible maternal genetic effects. An MR study by Moen et al. ^{Reference Moen, Brumpton and Willer80} investigating the association between maternal genotype and offspring adult disease, failed to identify any maternal genetic effects on offspring BW which were also causally associated with offspring cardiometabolic risk factors. In a following study, Warrington et al. ^{Reference Warrington, Beaumont and Horikoshi30} used MR to identify causality between maternal variants and offspring BW, independent of the fetal genotype. These analyses identified positive associations with offspring BW for maternal height and fasting glucose (through glucose level-associated SNPs), and negative associations for maternal SBP loci and maternal disposition index (a measure of beta cell efficiency) measurements. On the other hand, the MR analyses of fetal variants identified positive causal association with own BW for height alleles and fetal disposition index, but negative association for fasting glucose. In addition, no significant causal association driven only by the fetal genotype could be inferred for SBP loci. Building up on this study, Chen et al. ^{Reference Chen, Bacelis and Sole-Navais81} investigated the effect of parental genetics on offspring birth outcomes. Positive causal associations with offspring BW were found for maternal height, fasting glucose, and BMI. Analysis of the fetal genotype identified positive causal associations for height alleles (stronger than the maternal effect), and negative causal associations for fasting glucose and SBP. No significant fetal effect was found for BMI-related variants.

The Warrington et al. ^{Reference Warrington, Beaumont and Horikoshi30} and Chen et al. ^{Reference Chen, Bacelis and Sole-Navais81} studies produced very similar findings. The causal association between maternal BMI and higher offspring BW had been reported previously. ^{Reference Tyrrell, Richmond and Palmer82} On the other hand, the identified causal association between the maternal non-transmitted height genotype and offspring BW is in contrast to previous studies (albeit including a smaller sample size). ^{Reference Zhang, Bacelis and Lengyel83} The identified antagonistic causal effects on BW of maternal and fetal fasting glucose and disposition indexes (as a result from the maternal and fetal glucose metabolism genotypes) illustrate those early findings of Hattersley et al. ^{Reference Hattersley, Beards, Ballantyne, Appleton, Harvey and Ellard33} and support the fetal insulin hypothesis. ^{Reference Hattersley and Tooke36} Finally, there is a disagreement between both studies. Although Warrington et al. identified a maternal-origin only effect of SBP-influencing alleles, with no significant fetal effect, Chen et al. concluded that this relationship was driven exclusively by the fetal genotype. Further research is required to elucidate the extent of the effects of fetal and maternal SBP-related alleles on BW.

Future steps

The presented studies have identified nearly 200 loci associated with BW. Despite the advances in exploring the magnitude and source of the genetic influences governing BW, much work is still needed to properly identify, estimate, and comprehend these effects. The extent of the maternal genotypic influence on offspring development and later health outcomes needs to be determined, to overcome the confounding caused by its independent effect on growth and its correlation with the fetal genotype. Large and diverse cohorts with detailed neonatal phenotype and maternal and offspring genotype data are required to continue to decipher the influences of genetics on BW. Consortia such as the EGG or EAGLE ^{Reference Middeldorp, Felix, Mahajan and McCarthy84} pool results from multiple studies, with large numbers of individuals from different cohorts, to greatly increase the sample size and increase the power to identify genome-wide associations. The collaborative efforts of numerous research groups and institutions is therefore essential to provide the maximum number of individuals and insights to the cause. In addition, the availability of additional neonatal parameters could allow clarification of the genetic effects underlying BW variance. Gestational duration acts as a mediator of BW, but is not as commonly available as other neonatal measurements. Performing association analyses adjusted for gestational length would allow the identification of maternal genotype effects on offspring BW acting on the intrauterine environment and not simply through length of gestation. ^{Reference Warrington, Beaumont and Horikoshi30} However, this adjustment is not without drawbacks, introducing the risk of collider bias. ^{Reference Wilcox, Weinberg and Basso85}

Although maternal data are ubiquitous in studies determining parental correlation on offspring BW, very few studies include paternal information despite having a clear, though smaller, influence. ^{Reference Magnus20,Reference Griffiths, Dezateux and Cole86,Reference Rice and Thapar87} Studies including mother–father–offspring trios with available genotype data might shed more light on the way in which parental genes affect offspring BW. ^{Reference Hocher15} The paternal genotype might have the potential to influence offspring development not just by simple inheritance, but through active intervention in the intrauterine-environment via genomic imprinting. ^{Reference Fowden, Coan, Angiolini, Burton and Constancia88–Reference Herzog, Eggink and Willemsen91} Such studies will allow further exploration of how imprinted genes influence BW in terms of conflict between parental genomes. ^{Reference Haig92} In addition, studies have found associations between the neonatal epigenome (measured by DNA methylation) and BW. ^{Reference Madden, McCartney and Walker7,Reference Simpkin, Suderman and Gaunt93–Reference Küpers, Monnereau and Sharp95} Epigenome-wide association (EWA) analyses are promising and novel approaches to determine how genetics influence BW, investigating epigenetic gene regulation mechanisms and going beyond typical genotype analyses.

Finally, there is a severe lack of ancestry variety in the studies presented, with the individuals included in the GWA meta-analyses being predominantly European. In addition, most of these studies are based in high-income countries, and their results might not be readily generalizable to other populations, for example, those from low- or middle-income settings, which are subjected to different sets of environmental exposures. Future GWA initiatives should strive to include additional ancestral backgrounds, developing their studies in more diverse settings, to capture a wider set of environmental influences and genotypes and increase our understanding of how these two forces shape BW. ^{Reference Yokoyama, Jelenkovic and Hur96}