Discussion

This highly novel cohort study is one of very few human studies in the literature with examination data across three generations and both lineages which can be linked to children’s growth patterns at three different time points.22 To the best of our knowledge, there are no other reported human cross-generational data sets in the literature with four grandparental lines and three time points of measured BMI, so this is an original contribution to the scientific literature. This is also highly topical currently. Wells, in a recent commentary, indicates the need for such cross-generational grandparental data to understand more fully the influence of maternal and paternal lineages on growth and development.19

In a previous study employing primarily self-reported data, we showed that BMI appeared to be strongly familial in the maternal line,23 and we have shown that familial dietary patterns are also stronger in the maternal line.24 ,25 The present analysis of BMI has extended this to include both birth weight and data when the children were aged 9 years and shows this strong maternal pattern consistently for children while height, by comparison, is heritable across lineages. Similar patterns were seen across the generations for BMI and height. In our original publication, we used examination data at one time point only for the children when they were aged 5; we now report examination data at three separate time points, including a completely new sweep of data when children were aged 9.

It cannot be assumed that in a growing child these relationships with family members would remain consistently the same, in that environmental factors would come more into play as the child grows up.29 As children grow older in this study, the BMI relationship strengthens, presumably because of shared dietary environments, and to reinforce this, both grandparent couples also have significantly correlated BMI relationships. This is therefore a novel analysis and a considerable expansion on the previous one. In the original paper, we used only self-reported height and weight data for all adults, whereas this time we are reporting examination data for mothers and grandparents. Not alone is this now more robust, but also it is a substantial scientific addition using novel data to build on our previous observation. In this cohort, we have collected detailed information on social and environmental influences on the child's development and in a number of other publications address this question, including cultural and familial influences. We have, for instance, established dietary associations between nuclear family members and have shown that maternal pregnancy diet is associated with a number of outcomes, including child's risk of asthma, overweight and obesity.22

There are powerful environmental influences at play in familial studies. A report of the Young Finns Study followed offspring in relation to parental BMI from 3 to 18 years and later again when children were 24–39 years old. Maternal BMI was much more strongly associated with offspring BMI than that of father initially, but no differences in parental associations were seen when the offspring became adults.29

We approached this analysis looking for consistency and coherence of association as well as magnitude and statistical significance and include also the height analysis, which has the same numbers of respondents, to show that the absence of an association for BMI by comparison is not simply due to a power problem. While there are, for example, more mothers than fathers, there are relatively less grandparents, and yet both grandmothers do show statistically significant associations. Ponderal index is the clinical measure normally used at birth. However, we are not seeking to report or measure adiposity in itself in this analysis; for comparator purposes, we report BMI at the three time points in order to compare like with like both (1) in relation to the child over time and (2) in relation to the adults. Cole et al,30 for instance, concluded that ponderal index was not appropriate for measuring intrauterine malnutrition.

There were no correlations between BMI of the child of either sex at ages 0 and 5 with maternal or PGFs, so this evidence suggests that transmission of BMI is not primarily genetic. BMI of boys was correlated with that of the MGM at two points but was not correlated with that of the PGM. For the girls, there was a weak correlation with the MGM but a stronger correlation with the PGM at ages 0 and 5. The PGM transmits an X chromosome, which is the female sex chromosome, exclusively to her granddaughter, while girls on the maternal side may receive their X chromosome from either grandparent through their mother. The PGM does not transmit her X chromosome to the boys and the PGF does not transmit his Y chromosome to the girls. One of the highest p values for the girls (ie, the least significant correlation) was with the PGF, and the highest p values for the boys (ie, the least significant correlation) was with the PGM. Thus, this suggests that BMI is not transmitted via mitochondrial DNA or, if so, only weakly. This leaves the intrauterine environment or genetic transmission on the sex chromosome with evidence suggesting the latter, since there are no correlations with the grandfathers or of boys with the PGM.

A recent study17 found that female grandchildren have an increased risk of cardiovascular mortality if their PGM experienced a drastic change in food availability in their childhood. It indicated that X linked epigenetic inheritance via spermatozoa seemed to be plausible, with the transmission, limited to being through the father, possibly explained by the sex differences in the meiosis stage of cell reproduction. A similar explanation is plausible in this study for BMI.

We are not basing our conclusion that the maternal line is stronger solely on the fact that the mother–child correlations at age 5 are significant. We do so based on the facts that the overall maternal–child relationships are significant at all three time points, that there are correlations between all three mother–child pairs, for example, mother–child, MGM–mother and PGM–father, and that, in contrast to height, where both lineages show significant associations, the pattern for BMI is primarily the maternal lineage. Intrauterine transmission must be considered important as the maternal relationship is consistent at all time points and between each mother–child pair, for example, the mother and her children, the mother and her own mother and the father and his mother. While there have been conflicting reports in the literature with, for instance, the Avon Longitudinal Study of Parents and Children (ALSPAC)3 reporting equivalent relationships between the child’s BMI and both parents, the Generation R study2 reports that the maternal influence is strongest. A large Norwegian study31 has recently reported largely similar father–offspring and mother–offspring associations in adults across cardiovascular risk factors. However, in 9 of the 12 variables studied, age-standardised correlations were of a higher magnitude between mother and offspring, significantly so for height, systolic blood pressure, total cholesterol and high-density lipoprotein cholesterol. As has been pointed out,1 ,19 the relationship between the MGM in particular, her daughter and grandchild represents a 100 years of nutritional flow, including intrauterine, genetic, mitochondrial and environmental nutritional pathways and these data illustrate this effect. Morton et al16 indicate that population strategies aimed at optimising size at birth require intergenerational considerations for social as well as biological reasons.

One key reason we conclude that intrauterine or X-mediated influences are important is because the grandmothers of both lineages, who do not share their grandchildren's direct daily environment, are each exerting an association. Most studies do not have grandparental data from both lineages. The MGM could arguably have influenced her daughter's culinary and dietary habits and hence her grandchild's BMI,25 but the PGM is associated with the paternal granddaughter and not her paternal grandson. It is entirely plausible that this is biologically mediated, and not purely environmental, as we posit, and was also reported in the Overkalix study.17 ,18

In a meta-analysis of 26 family studies,5 BMI heritability estimates ranging from 0.24 to 0.81 were reported. Our correlation estimates (corresponding to half the heritability) of 0.25 between the mother and the MGM and 0.26 for the father and the PGM are consistent with these as is the value 0.28 for that of the child at age 9 with the mother. The analysis5 reported that heritability estimates rose by 0.012/year throughout childhood (age ≤18 years) but decreased by −0.002/year in adulthood. Our correlation estimates of 0.15 of the child at age 0 with the mother rising to 0.25 at age 9 and averaging at 0.25 in adulthood is consistent with this with the caveat that no change was seen between age 0 and 5. Note that a similar pattern was not seen for offspring and fathers. Also note that, as in that analysis,5 we found no difference in BMI heritability estimates between men and women. In contrast to BMI, height correlation estimates between the offspring and the mother are quite similar to those between the offspring and the father for each of the three child ages and each generation. The estimates increased rapidly from age 0 to 5 and then at a slower rate, reaching maximum values in adulthood. The estimates are greater than those for BMI throughout.

Modern techniques for genetic analysis create opportunities for understanding the heritability of complex traits and risk for chronic diseases.27 ,32 Height is one of the more complex genetically inherited traits because as many as 20 genes—if not more—interact to determine a child’s height. Geneticists have found genes for height on the 7th, 8th and 20th chromosomes, as well as on the X chromosome. A recent report shows that combined common single-nucleotide polymorphisms explain a large proportion of the heritability for human height.33 What is under-researched to date is the degree to which the X chromosome might be contributing to human obesity.34 Unlike other authors,35 we did not find heritability for postnatal length to be higher for maternal height than for paternal height. In our analysis, it was found that the child’s height was correlated with that of all family members at age 5, suggesting perhaps many genetic effects. Thus, the common estimate of heritability assuming additive genetic effects may not work for height.36 However, at ages 0 and 9, height was primarily correlated in the nuclear family and similar correlations were observed for the parents with the grandparents. Interestingly, the height of the child continued at ages 5 and 9 to be strongly correlated with that of the PGF, not easily explainable by conventional genetic inheritance, though there may be a Y-related influence at play through the father, who himself is highly correlated with his father.

There was a systematic and statistically significant difference in the height of the two groups of fathers—those reported were significantly taller than those measured. This may be because reported data were exaggerated by the partner or because, for some reason, reported is a healthier group than measured height. For the small numbers of fathers we measured, there was a significantly lower height than in those we reported. However, the latter explanation is less likely since the two groups do not differ in weight or heights and weights of their partners—that is, the mother, or heights and lengths of child at age 0. It is well known that male heights tend to be greater in reported than measured data.37 Our analysis suggests that it is unlikely that correlations were affected by this feature.

Probably the most familiar technique in estimating heritability is parent–offspring regression in which, for example, a trait's heritability can be estimated as the slope of the regression of offspring phenotype on mid-parent phenotype. It has been noted36 that if the sexes are not equal in phenotypic variance, then regression of daughters on fathers, for example, must be adjusted for the difference in variation. The adjusted regression is the correlation between the phenotypes. Twice this correlation is then defined to be the heritability. Thus, correlations between the phenotypes reported here can be taken to approximate half the heritability. In addition, adjustments for multiple testing were made in this model. Studies on children are likely to be over-representative of individuals from more recent birth years, but the separation of birth cohort and age effects on BMI correlation was possible in this study by including age-fixed and generation-fixed effects in the model. Age–genotype interactions are also modelled by the latter. Thus, correlation estimates obtained are comparable across cohorts and age groups.

Our data do have limitations, as they are based on a relatively small-scale cohort with differential response rates at each sweep and the numbers of measured children at age 9 was considerably lower than in the previous sweeps. We agree that caution is appropriate accordingly in data interpretation as the numbers at age 9 are smaller. In previous publications, we have reported on the representativeness of the participating mothers and children and find, as with many cohorts, that less affluent and younger mothers were less likely to continue to participate, but the baseline maternal BMI does not differ between responders and non-responders.22 Also, we only report statistically significant findings which by definition are sufficiently powered when the child is aged 9. There may be some self-selection bias in relation to participating grandparents, as clearly grandparents who were deceased could not participate.26 In previous analyses, we have shown that the original index mothers were representative of the general population,22 and our findings were consistent across the three sweeps, so we do not think self-selection can be the main explanatory factor for our findings.

In conclusion, this novel cohort study shows clear familial associations across three generations for height in both lineages, but the BMI pattern is stronger in the maternal line.