^{1}

^{2}

^{1}

^{1}

^{2}

^{*}

^{1}

^{2}

Edited by:

Reviewed by:

*Correspondence:

This article was submitted to Applied Genetic Epidemiology, a section of the journal Frontiers in Genetics.

This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

Maternal genetic and phenotypic characteristics (e.g., metabolic and behavioral) affect both the intrauterine milieu and lifelong health trajectories of their fetuses. Yet at the same time, fetal genotype may affect processes that alter pre and postnatal maternal physiology, and the subsequent health of both fetus and mother. We refer to these latter effects as ‘fetal drive.’ If fetal genotype is driving physiologic, metabolic, and behavioral phenotypic changes in the mother, there is a possibility of differential effects with different fetal genomes inducing different long-term effects on both maternal and fetal health, mediated through intrauterine environment. This proposed mechanistic path remains largely unexamined and untested. In this study, we offer a statistical method to rigorously test this hypothesis and make causal inferences in humans by relying on the (conditional) randomization inherent in the process of meiosis. For illustration, we apply this method to a dataset from the Framingham Heart Study.

A common belief is that genetic, phenotypic, and behavioral characteristics of mothers affect the metabolic milieu during pregnancy, and that this in turn affects both the short and long-term metabolic health of the offspring, including the risk of obesity. This is likely true and some evidence is supportive (

The maternal–fetal relationship can and has been viewed as both one of intimate cooperation and that of intergenerational competition (

Identifying specific aspects of fetal genotypes (beyond the differential effects of sex chromosomes) influence maternal phenotype may be challenging for all the reasons that studying complex genetic effects are challenging in general, but also because a statistical model for testing such hypothesized effects has not yet been offered. Except in certain situations of controlled experimental crosses, observational studies showing an association between fetal genotype at a particular genomic locus and maternal phenotype merely indicate association (correlation) and not necessarily of causation. Therefore, we offer a statistical model, adapted from one developed by one of the authors in the context of family based association (i.e., ‘TDT’) testing, to rigorously test this hypothesis in humans (and other diploid populations) and make causal inferences by relying on the randomization inherent in the process of meiosis. Although the model is adapted from genetic study, the principle has its theoretic root in the causal inference literature. We describe the method here and illustrate it with data from the Framingham Heart Study (FHS).

The essential proposition underlying the validity of the method we offer is that, under Mendelian theory, at any genetic locus, conditional on the parents’ mating types at that locus, the offspring’s genotype at that locus is a random variable for which the probability of each possible genotype is equal for all offspring. Therefore when we condition on parents’ mating types at a locus of interest, then we have effectively a randomized experiment in which offspring are randomly assigned to genotypes at that locus. Stated, equivalently, mothers are randomized to carry offspring of different genotypes at that locus. Hence, if we test for relations of offspring genotype with mother’s phenotype during pregnancy (e.g., preeclampsia, gestational diabetes, weight gain, or for that matter any time after conception), we can reasonably draw causal inferences about the effects of fetus genotype on mother’s pregnancy because of the aforementioned randomization,

The above principle has been formalized in the literature of causal inference (_{F},g_{M}), are the _{k}, is the treatment although we have three groups here (i.e., three types of genotype); mother’s phenotype is the dependent variable

where _{k} is genotype of fetus, m is the parental mating types at the same locus, ε is random error. In order to test the hypothesis, we only need to test H_{0} : β_{k} = 0. Note that the genotypes should be treated as categorical in the model (

For illustration, we applied the concept and the model to a dataset from the FHS. The study began in 1948 with 5,209 adult subjects (i.e., the first generation) from Framingham, Massachusetts, and is now on its third generation of participants. The Offspring Study (i.e., the second generation) was initiated in 1971. A sample of 5,124 men and women, consisting of the offspring of the Original Cohort and their spouses was recruited. The recruitment of the third generation participants who had at least one parent in the Offspring Study and would be at least 20 years old by the close of the first exam cycle was started in 2001. A recruitment target of 4,095 Gen III participants was achieved by July of 2005. In order to draw causal inference using the model as described above, we need information from both parents and offspring (although extensions to allow for missing data from one parent with substitution of sibling data are possible (

Although information on gestational hypertension, diabetes, and weight gain is available in FHS, in the final dataset we obtained, the number of observations of hypertension and diabetes are small. Therefore we only focus on weight gain, which is binary with value 1 if the mother reported to have >30 pounds weight gain during pregnancy and 0 otherwise, that is:

where I( ) is the indicator function.

FTO is a gene located in chromosome region 16q12.2. Studies have revealed association of single nucleotide polymorphisms (SNPs) in this gene with obesity. The genotype we tested are four SNPs (i.e., rs9930506, rs9939609, rs1121980, and rs8050136) in FTO gene which have been shown to be associated with obesity (

To analyze the data, we used logistic regression. We performed analyses including and excluding mother’s age at children’s birth in the model. For the analyses we conducted, none of the mother’s age at children’s birth was significant (at α = 0.05). Therefore they were not included in the final models.

Results from logistic regression for SNP rs9930506 from data with only the first child in each family.

Independent variable | Intercept | Fetal genotype | Mating type | |||||
---|---|---|---|---|---|---|---|---|

Coefficient | –1.504 | –0.049 | 0.239 | 0.177 | –0.527 | 0.200 | 0.860 | 0.133 |

0.054 | 0.928 | 0.974 |

Statistical model | SNPs |
|||
---|---|---|---|---|

rs9930506 | rs9939609 | rs1121980 | rs8050136 | |

Logistic regression | 0.928 | 0.946 | 0.944 | 0.965 |

The hypothesis that the fetal genotype affects maternal physiology and behavior represents a complementary view about not only some gestational diseases (e.g., pregnancy-induced hypertension and gestational diabetes) of mothers, but also long-term effects on metabolic health of both mother and offspring. In addition, the effects of this process may be cyclic over generations (

Both animal studies and human studies have shown the effect of fetal genotype on mother’s physiology during pregnancy (

It should be noted that random mating is not assumed for our model. The model is based on the randomization of offspring via the random assignment to genotypes at the locus under study. The randomization holds

Our model is based on a linear model (more general, generalized linear model) and assumes samples with unrelated study units. The model can be extended to handle data with related study units, such as biological siblings, by the use of a linear mixed model (or more general, generalized linear mixed model). The correlation between the related individuals can be dealt with via a random effect by using a matrix of kinship coefficients (a kinship coefficient is a measure of degree of genetic correlation between two individuals;

There are no significant findings in our data analysis. This may be because that the four SNPs we tested did not have deleterious effects on maternal metabolic function, or our analysis lacked the requisite power to detect an effect. Statistical power is influenced by many factors including effect size and sample size. Metabolic function is a complex phenotype, and we did not expect a large effect size for any of its risk factors. We only tested four SNPs in FTO gene due to its relationship with obesity, in a small sample, and think that a genome-wide scan in a much larger sample is warranted for more reliable inferences.

Although our statistical method makes the direct test of our hypothesis possible in human studies, it should be noted that conducting such studies necessitate a great deal of time and effort because both parents and offspring need to be included in the study and relevant physiological measurements must be made in the mothers during pregnancy. The effort is worthwhile because if the hypothesis is validated, our understanding of some gestational metabolic conditions may be shifted to a new level. New strategies may be developed to prevent and reduce morbidity and possibly mortality in both the mother and offspring and their future descendants.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Supported by NIH grants P30DK056336, T32DK062710, R25DK099080, R01DK052431, R01GM081488, R01HL092173, S10RR026723, and P60AR064172.