Introduction
It is now well established that infants’ birth weight is associated with longer term morbidity and mortality, both overall and specifically in relation to cardiovascular disease.Reference Osmond and Barker1, Reference Whincup, Kaye and Owen2 The causal pathways remain to be elucidated, but several possible mechanisms have been proposed. Processes occurring during intrauterine foetal development may programme cardiovascular development through possible epigenetic and nutri-genomic mechanisms.Reference Hanson, Godfrey, Lillycrop, Burdge and Gluckman3 Common genetic factors may influence both foetal growth and later morbidity in adulthood.Reference Hattersley and Tooke4 It is also possible that environmental factors, including social position, influence these outcomes.Reference Bergvall and Cnattingius5 Familial studies can contribute to this investigation, as shared heritable and environmental conditions occur. There is now mounting evidence that mothers of low birth weight (LBW) babies subsequently have higher rates of cardiovascular morbidity (CVD) and mortality,Reference Friedlander, Paltiel and Manor6–Reference Davey Smith, Sterne, Tynelius and Rasmussen16 which is concluded as evidence for an intergenerational intrauterine effect consistent with the foetal origins hypothesis.Reference Davey Smith, Hypponen, Power and Lawlor15 This means it is not just LBW individuals who run long-term risk themselves, but the mothers of such infants as well. What is particularly intriguing is whether and how these mechanisms are transmitted across generations and whether these follow patterns according to sex and type of familial relationship, which constitutes the research questions for this study.
An infant shares half, respectively, of each parent's genes, and usually in common its early childhood environment. Parent–child studies have shown a stronger association for mother–infant birth weight than father–infant birth weight association.Reference Hennessy and Alberman17, Reference Horta, Gigante, Osmond, Barros and Victora18 Familial studies also suggest that behaviours such as dietReference Brion, Ness and Rogers19 and smokingReference Al Mamun, O'Callaghan and Alati20 are transmitted down familial lines with a specific maternal intrauterine influence. The infant's relationship with its mother is important, both because of the intrauterine exposure experience and mitochondrial heritability through ovum.Reference McConnell21
Animal and human studies have indicated possibilities of both genomic and non-genomic mechanisms in intergeneration transmission of disease risk.Reference Drake and Walker22, Reference Drake and Liu23, Reference Gluckman, Hanson and Beedle24 If pathways are of genetic or environmental continuity, one might expect similar familial patterns in both lineages across more than one generation. On the contrary, if pathways are intrauterine one might expect the associations to differ in lineages, as additionally during the critical stage of in utero development mothers are likely to pass such information to foetus, which helps in aligning it to the phenotype of matrilineal ancestry.Reference Kuzawa25, Reference Kuzawa26 However, grandparental transmission pathways have not been well studied to date. There are just four accounts of the relative influence of the grandparents’ health status on their grandchildren's birth weight and these are all through retrospective cohort or linkage designs.Reference Smith, Wood, White, Pell and Hattie27–Reference Naess, Hoff and Lie30
Smith et al.Reference Smith, Wood, White, Pell and Hattie27 in a large Scottish linkage study showed a pattern of inverse risk of ischaemic heart disease and cerebrovascular diseases in maternal grandparents related to their grandchildren's birth weight. Adjustment for maternal characteristics, including social position, reduced the effect but associations remained significant. This study was restricted only to maternal grandparents.
Manor and KoupilReference Manor and Koupil28 independently examined all four grandparents and showed varied associations between grandparental mortality and birth weight of their grandchildren in an Uppsala birth cohort linkage study. They demonstrated a U-shaped association for maternal grandfathers’ (MGF) overall and circulatory disease mortality, and also for maternal grandmothers’ (MGM) circulatory disease mortality. They also showed inverse associations for paternal grandmothers’ (PGM) circulatory disease mortality, and paternal grandfathers’ (PGF) overall and circulatory disease mortality. Adjustment for social position made no changes to observed associations.
McCarron et al.Reference McCarron, Davey Smith and Hattersley29 reported from the ALSPAC cohort a difference in the mean birth weight of grandchildren with at least one diabetic maternal grandparent compared with those with at least one diabetic paternal grandparent or no diabetic grandparent. Grandchildren of diabetic maternal grandparents were on the highest tertile of birth weight compared with non-diabetics; they also suggested an inverted U-shaped association for diabetic paternal grandparents, which was not statistically significant and attenuated with adjustments.
Naess et al.Reference Naess, Hoff and Lie30 in a large Norwegian linkage cohort identified that history of myocardial infarction in MGMs was related to infant birth weight but no effect was seen for other grandparents. The authors of these studies hypothesized that some common mechanism explains both cardiovascular risk in grandparents and birth weight in infants.Reference Smith, Wood, White, Pell and Hattie27 The grandmaternal association was generally interpreted as an intrauterine effect.Reference Manor and Koupil28, Reference McCarron, Davey Smith and Hattersley29, Reference Naess, Hoff and Lie30 The two studiesReference Manor and Koupil28, Reference McCarron, Davey Smith and Hattersley29 with paternal grandparental findings suggested a possible epigenetic X-linked mechanism as an explanation of their varied results.
The objectives of this analysis employing data from the Lifeways cross-generation cohort study of a thousand Irish families (Lifeways)Reference O'Mahony, Fallon and Hannon31 were to assess whether grandparents’ cardiovascular risk factors, morbidity and mortality related to their grandchildren's birth weight, and if so whether there were lineage and gender-specific associations. The study intended to replicate the analysis strategy of previously published grandparental studies.Reference Smith, Wood, White, Pell and Hattie27, Reference Manor and Koupil28, Reference McCarron, Davey Smith and Hattersley29, Reference Pembrey, Bygren and Kaati32
Method
Lifeways was established in Ireland in 2001 with the a priori purpose of examining how familial and cross-generation factors influenced early childhood growth and development and its recruitment procedures have been described previously.Reference O'Mahony, Fallon and Hannon31, Reference Kelleher, Fallon and Fitzsimon33 Briefly, mothers were recruited at first hospital booking visits in one of two regional maternity hospitals in Dublin and Galway over a 15-month period. Mothers were asked to recruit other family members including the partner and if possible at least one grandparent. Of 1094 live births, including 12 twin pairs, 610 of 1082 families had at least one participating grandparent. Families with additionally available infants’ gestational age information totalled 539. To maximize the number of available families, rather than exclude the twin families, data on the same-sex twin infants were combined to give one average birth weight, a procedure employed previously in similar studies with multiple children per parent/grandparent.Reference Davey Smith, Hart and Ferrell14, Reference Manor and Koupil28 Two families with mixed gender twins were excluded as birth weight of male and female twins may differ. In the final analysis sample, there were just 8 twin families. There were only few (40 grandparents from 30 families) formal withdrawals. Thus, the study sample comprised 539 families (539 infants, 1054 grandparents). Figure 1 provides a schematic presentation of the study.
Fig. 1 Schematic presentation of the information available in the study. BMI, body mass index; CVD, cardiovascular disease.
The cohort was previously reported to be representative of the general Irish population.Reference Niedhammer, O'Mahony, Daly, Morrison and Kelleher34 However, there were some socio-demographic differences of families with participating grandparents, compared with those without, shown in Table 1. In this cohort, a social gradient to birth weight and the degree to which pregnancy outcome was explained by lifestyle factors including diet has also been previously published. As expected, maternal smoking was an important predictor of birth outcomes in these studies, as were factors such as living conditions.Reference Niedhammer, Murrin and O'Mahony35, Reference Murrin, Segonds-Pichon and Fallon36 Despite these observed social differences, there were no differences in biological variables such as height and BMI of mothers and grandparents in family groups that either participated or not in the analyses of present study (Table 1).
Table 1 Socio-anthropometric profile of Lifeways families
GMS, General Medical Services (means tested healthcare); BMI, body mass index.
*P < 0.05 (ANOVA and χ 2).
Information on grandparents was available at three different time points. (1) At baseline (recruitment/the antenatal stage for probands), grandparents were asked to complete a self-administered questionnaire, which included questions on diagnosed cardiovascular status.Reference Kelleher, Fallon and Fitzsimon33 They were also offered a cardiovascular risk factor assessment at home, conducted by trained nurses. (2) When the grandchildren averaged 3 years old, in 2005, with informed consent from adult cohort members, a note search was undertaken in their general practice, which recorded updated status on cardiovascular diagnosis. (3) In the interval between recruitment and 2010, family members were maintained in regular contact and asked to provide information on major inter-current illness or death of grandparents. In 2010, a formal search was made in the General Register Office for death certificates using grandparents’ name, age and address.
We thus performed three analyses relating infants’ birth weight to grandparents’ health status. First, we looked at the relationship between the grandparents’ cardiovascular risk factor profile at recruitment, the antenatal stage for probands and grandchildren's birth weight. We then examined the pattern of relationship between grandparental disease-specific morbidity and grandchildren's birth weight using a composite measure of any diagnosed CVD from questionnaire, risk factor assessment or general practice records. Cardiovascular risk factors were available for 79% of grandparents in these families and general practice follow-up records for 61%. Finally, we examined the grandparents’ mortality pattern, looking at the confirmed deaths at 10-year follow-up in all 539 families.
1. Risk factor examination of grandparents
At baseline, 79% grandparents participated in a cardiovascular risk assessment (European Health Risk Monitoring protocol, http://www.ktl.fi/ehrm/) inclusive of anthropometric [height (cm), weight (kg), waist circumference (cm), hip circumference (cm)], blood pressure [three readings each of systolic and diastolic blood pressures (mmHg)] and five component serum lipid profile (mmol/l) assessments.Reference Kelleher, Fallon and Fitzsimon33 Overall, 828 grandparents from 427 families that had measured cardiovascular risk factors and subsequent birth of a live healthy proband (grandchild) in their families with available gestational age information were analysed. Initially, clinical predictors of infants’ birth weight (g) were examined (partial correlations) with all grandparents clustered together, followed by separate examination by lineage and four grandparent–grandchild dyads. Statistically significant parameters were re-examined in multivariable linear regression models, adjusted as appropriate for children's characteristics (gestational age, gender), grandparents’ characteristics (age, smoking, height, waist–hip ratio) and maternal prenatal characteristics [age, parity, smoking, body mass index (BMI)]. Finally, the significant predictors were further analysed by grandchildren's gender to understand gender-specific associations.
2. Measures of CVD
At baseline, diagnosed status on stroke and diabetes was available for grandparents from almost 539 families (1054 grandparents) that had live healthy grandchild and available gestational age data, but history of myocardial infarction was classifiable only from the self-reported questionnaire (n = 364 families). In 2005, general practitioners provided an update on CVD status for 321 out of 539 families (61% grandparents). Initially for all grandparents clustered together, the odds of grandparents having cardiovascular diseases, per 100 g increment in grandchildren's birth weight, were examined using logistic regression. The difference in adjusted mean birth weight (g) of grandchildren between disease reporting and non-disease reporting grandparents were then examined by ANCOVA for lineage and four independent grandparent subgroups. Logistic regression and ANCOVA models were adjusted for children's characteristics (gestational age, gender), grandparents’ characteristics (age, smoking, height, BMI) and maternal prenatal characteristics (BMI).
3. Prospective follow-up for mortality 2010
In 2010, the death registry search at the General Register Office confirmed 85 grandparental deaths since baseline in 610 families, 77 of these in the 539 families (1054 grandparents) with infant gestational age information. Hazard ratios (HR) for all-cause mortality were calculated with Cox proportional hazards regression. As the samples were small, we focussed on only all-cause mortality. HR for grandparents’ mortality were compared for LBW and high birth weight (HBW) infants against normal birth weight infants. LBW was defined at both clinically relevant cut-off <2500 gReference Friedlander, Paltiel and Manor6, Reference Friedlander, Manor and Paltiel12 and epidemiologically relevant cut-off <3000 g.Reference Whincup, Kaye and Owen2 HBW was defined as ⩾4000 g.Reference Friedlander, Paltiel and Manor6, Reference Friedlander, Manor and Paltiel12 HR was also examined with birth weight (g) as a continuous independent variable.Reference Davey Smith, Hart and Ferrell14, Reference Smith, Wood, White, Pell and Hattie27 Time scale was duration in months elapsed since birth of grandchildren to either censoring of study (15 November, 2010) or death of grandparents. Age of grandparents at the entry point of study (birth of grandchildren) was adjusted as covariate in all models. Models were adjusted for children's characteristics (gestational age, gender), grandparents’ characteristics (age, smoking, education) and maternal prenatal characteristics (age, parity, smoking, height, education).
To replicate the analytical strategy taken by McCarron et al.,Reference McCarron, Davey Smith and Hattersley29 at univariate levels, the grandchildren's standardized birth weights (standardized for gestational age and gender using linear regressions)Reference McCarron, Davey Smith and Hattersley29 were compared in family groups according to grandparental status of experiencing specific CVD or all-cause mortality. Analogous to their approach, analyses were conducted at isolated family group levels, that is, at least one maternal grandparent only, at least one paternal grandparent only and then one by one for each of the four grandparents. At each level, whereas the case families were families that experienced the specific cardiovascular event or mortality of only the grandparent in context, the control families were families that did not. Thus, the case and control families were alike except for the birth weight influence generated by health event of the concerned grandparent in context. This prevented dilution of any birth weight effect that grandparent of one type may generate by effect generated by grandparent of another type. An additional control group was families where no health events for any type grandparents (‘None grandparents’) was observed.Reference McCarron, Davey Smith and Hattersley29 Grandparental lineage differences could also be judged by comparing mean infant birth weight values in case of family groups for any maternal grandparent only v. any paternal grandparent only.Reference McCarron, Davey Smith and Hattersley29
In familial analyses, the same event can be examined as an outcome or exposure.Reference Smith, Wood, White, Pell and Hattie27, Reference Susser and Susser37 First, the excess risk of birth weight (outcome) for the proband (grandchildren) was examined, considering cardiovascular risk factors and disease status of their grandparents as exposures. Next, the excess risk of all-cause mortality (outcome) was examined for the grandparents, considering the birth weight of their grandchildren as an exposure. As the study was longitudinally designed, we preferred analysing the events in temporal sequence.
The statistical analyses were run on four levels: (i) All grandparents – all grandchildren level, which tested cross-generational associations independent of the lineages. In these models, additional adjustments were made for grandparents’ maternal–paternal lineage and gender. (ii) Maternal and paternal lineage comparisons were made by analysing associations for families, wherein any of the maternal grandparents or any of the paternal grandparents reported to be affected by the condition in context. This was achieved by replicating the analytical approach taken by Smith et al.,Reference Smith, Wood, White, Pell and Hattie27 independently coupling both the grandparents with the same grandchild by clustering on grandparental identifier, instead of family identifier. These analyses were additionally adjusted for the gender of the grandparent. (iii) Four different types of grandparent–grandchild dyads. (iv) Sex-specific associations by separately testing grandparent–grandchild dyads according to grandchildren's gender. This generated eight sex-specific pairs for analyses. The rationale for testing lineage, four grandparental and gender-specific levels has been stated earlier. If each grandparent–grandchild (grandson/granddaughter) has a specific relationship, pooled analyses may lead to false conclusions.
The grandchildren's birth weights were within plausible limits. In accordance with the approach taken by previous researchers, only live births were considered for analysis and stillbirths or lethal congenital anomalies were excluded.Reference Friedlander, Paltiel and Manor6, Reference Pell, Smith and Wash9, Reference Smith, Pell and Walsh10, Reference Manor and Koupil28 As there were no systematic mechanisms that explained missing data, it was dealt as missing completely at random using complete case analysis approach, which is listwise deletion of families with missing data on any of the confounding variables.Reference Howell38 All analyses were performed using SPSS version 15.0 (SPSS Inc., Chicago, IL, USA). Children's birth weight, gestational age, gender and date of birth, as well as mothers’ date of birth and parity, were determined from hospital records. Gestational age at birth was computed from estimated date of delivery and actual date of birth of the infant. Maternal and grandparental age was defined as age at birth of the grandchild. Parity was defined as previous live or dead births after 28 weeks of gestation. Maternal smoking was her smoking status at first antenatal care visit. Grandparental smoking was their smoking status at recruitment. Maternal and grandparental educational levels were used as proxy for familial socio-economic status.
Results
(1) Association with cardiovascular risk factors – antenatal stage
The mean values of the measured cardiovascular clinical parameters are given in Table 2; no differences were observed in the risk profile of grandparents by lineages.
Table 2 Mean values of CVD clinical parameters in grandparents (n = 828 grandparents from 427 families)
CVD, cardiovascular disease; MGM/MGF, maternal grandmother/father; PGM/PGF, paternal grandmother/father; HDL, high-density lipoprotein; LDL, low-density lipoprotein; BP, blood pressure; BMI, body mass index.
Initial analyses of CVD risk factors only suggested a weak negative correlation with all grandparents’ high-density lipoprotein cholesterol (HDL cholesterol) and grandchildren's birth weight (r = −0.1; P < 0.01), not attenuated by adjustments. However, independent grandparent–grandchild dyad linear regression analyses (Table 3) revealed that PGMs’ triglycerides [unstandardized-β (95% CI) = 78.8 (7.0 – 150.7); P = 0.03] and PGFs’ systolic blood pressure [unstandardized-β (95% CI) = 6.6 (0.8 – 12.5); P = 0.03] were the only risk factors consistently predictive of grandchildren's birth weight. PGMs also showed a positive association for systolic blood pressure, but only after adjustments. Although no pattern was seen for MGFs, an association observed for MGMs’ diastolic blood pressure attenuated with adjustments. The associations for both paternal grandparents only got stronger on adjustments, and social position did not weaken these associations.
Table 3 Grandparents’ cardiovascular risk factors – grandchildren's birth weight associations for grandparent–grandchild dyads: birth weight (g) regressed on clinical parameters (n = 828 grandparents from 427 families)
BP, blood pressure; HDL, high-density lipoprotein; MGM/MGF, maternal grandmother/father; PGM/PGF, paternal grandmother/father; BMI, body mass index.
β 1: unstandardized β coefficient; *P < 0.05.
M 1: Model 1: adjusted for grandchildren's gender, gestational age and grandparents’ age, smoking.
M 2: Model 2: Model 1 plus adjusted for grandparents’ height, waist : hip ratio, education and mother's age, parity, smoking during pregnancy, pre-pregnancy BMI, education.
Mx: Sex-specific model: adjusted for only following variables: grandchildren's gestational age; grandparents’ age, smoking, height; and mother's pre-pregnancy BMI.
Gender-specific analyses according to sex of both grandchildren and grandparents (Table 3) conducted to test sex-specific trans-generational responseReference Pembrey, Bygren and Kaati32 again showed a number of significant associations. However, the strongest and most consistent association was that observed for PGMs’ triglycerides [unstandardized-β (95% CI) = 128.2 (14.8 – 241.6); P = 0.03] with birth weight of granddaughters. During analysis, all models for triglycerides were statistically significant and the strength of the relationship increased in an ordered pattern with adjustments for confounders, suggesting that this association was not by chance.
(2) Association with established CVD – antenatal stage and a 3-year follow-up
Twenty-five cases of myocardial infarction were reported. Infants’ birth weights were inversely associated with all grandparents’ reported myocardial infarction [OR (95% CI) = 0.88 (0.79 – 0.98); P = 0.02] but numbers were too limited to examine this further. The associations were not significant for stroke and diabetes at all grandparents’ level.
Initially in Table 4 univariate comparisons are shown for mean standardized (for gestational age and gender) birth weights in grandchildren by grandparental status of stroke or diabetes. Subsequently, the difference in grandchildren's birth weight after complete adjustments for child, maternal and grandparental characteristics using ANCOVA (β-values) are presented.
Table 4 Difference in grandchildren's mean standardized BW by grandparents’ stroke status (n = 1035 grandparents from 537 families) and grandparents’ diabetes mellitus status (n = 1044 grandparents from 536 families)
ref, reference group; BW, birth weight; MGM/MGF, maternal grandmother/father; PGM/PGF, paternal grandmother/father.
aBWs (g) standardized for children's gestational age and gender (using linear regression).
bGrandparent had stroke/diabetes mellitus.
cGrandparent did not have stroke/diabetes mellitus.
dDifference in grandchildren's adjusted means of BW (g) if grandparents had stroke/diabetes (ANCOVA), adjusted for children's gender, gestational age; grandparents’ age, smoking, height, BMI; and mothers’ pre-pregnancy BMI.
§P < 0.1; *P < 0.05; **P < 0.01.
In univariate comparisons (Table 4a–4b), the infant birth weights for affected maternal grandparents were lower than those of unaffected maternal or all (‘None’) grandparents, but conversely the infant birth weights for affected paternal grandparents were higher than non-affected paternal or all (‘None’) grandparents. Similarly, the infant birth weights of affected maternal grandparents were lower than affected paternal grandparents. This diverging pattern was also true for independently tested four grandparent–grandchild dyad groups, except for MGFs in respect to stroke.
Results from ANCOVA analysis demonstrate that fully adjusted mean birth weight for infants of MGMs, who were diagnosed with stroke (Table 4a), or maternal grandparents, who were diagnosed with diabetes (Table 4b), remained significantly lower than those without. In contrast, paternal grandparents’ reported stroke or diabetes seemed associated with infants’ higher birth weight, although not significantly so. The status of both stroke and diabetes in MGMs was significantly inversely associated with infants’ birth weight (Table 4a–4b), and this remained so even with adjustment for social position. There was also a borderline significant inverse pattern for MGFs’ diabetes status and infants’ birth weight (Table 4b).
(3) Association with all-cause mortality – a 10-year follow-up
In Table 5, we examine the grandparental mortality pattern. In univariate comparisons (Table 5a), a similar pattern was observed. Whilst the grandchildren's mean birth weights for deceased maternal grandparents were lower than living maternal or all (‘alive’) grandparents, the grandchildren's birth weights for deceased paternal grandparents were higher than living paternal or all (‘alive’) grandparents. Again the infant birth weights of deceased maternal grandparents were lower than deceased paternal grandparents. This contrasting pattern was also observed in independent examination of MGMs, MGFs and PGFs, with the difference for the PGFs being significantly large.
Table 5 (a) Grandchildren's mean standardized BW by grandparents’ mortality status and (b) HR for grandparents’ death in groups defined by grandchildren's BW status (n = 1054 grandparents from 539 families)
BW, birth weight; HR, hazard ratio; LBW, low birth weight; HBW, high birth weight; ref, reference group; MGM/MGF, maternal grandmother/father; PGM/PGF, paternal grandmother/father; NA, inadequate information for analysis (no deaths).
aBW groups in HR results are independent and not ordered.
bReference is 2500–3999 g.
cReference is 3000–3999 g.
dNumber deaths after excluding other dead grandparents.
eBWs (g) standardized for children's gestational age and gender (using linear regression).
fGrandparent died.
gGrandparent alive.
HR adjusted for grandchildren's gender, gestational age; grandparents’ smoking, educational status, age at grandchildren's birth; and mother's age, height, parity, educational status, smoking during pregnancy.
*P < 0.05; **P < 0.01.
Cox proportional hazard models (Table 5b) demonstrate an association between lower birth weight infants (both <2500 g and <3000 g) and grandparental mortality only in maternal line families, although not significantly so. A U-shaped association with MGMs’ mortality was consistently observed, but was not statistically significant, whether adjusted or not for maternal and grandparental characteristics. In contrast, a direct significant relationship emerged between PGFs’ mortality and higher birth weight infants (>4000 g), which strengthened on controlling for maternal and grandparental characteristics, inclusive of smoking and social position. When analysis was repeated using birth weight (g) as a continuous independent variable, this association for PGFs was found to be linear [adjusted-HR (95% CI) = 1.002 (1.000 − 1.003); P = 0.04].
Discussion
This analysis shows that there is a pattern of association between grandparents’ health status and the birth weight of their grandchildren at three separate time points: in the year the child was born, at follow-up through general practice three years later and at prospective follow-up a decade after recruitment. As the Lifeways cohort study was designed with the a priori objective of assessing cross-generational influences on early-life risk, the present study has an advantage in comparison to those published previously on this topic. In addition, to our knowledge it appears to be the first to relate measured grandparental cardiovascular risk factors, as opposed to the only reported information, to grandchildren's birth weight.
Although the literature on parental risk factors, morbidity and mortality is now sizeable,Reference Friedlander, Paltiel and Manor6–Reference Davey Smith, Sterne, Tynelius and Rasmussen16 there is very little published to date on cross-generational patterns.Reference Smith, Wood, White, Pell and Hattie27–Reference Naess, Hoff and Lie30 In addition, lineage and gender-specific patterns have not been contrasted. Our analyses demonstrate consistently different association patterns in maternal and paternal lines. Both ends of the birth weight spectrum have been examined. The association with high birth weight is observed for the paternal line, in the case of PGFs’ mortality and positively with unfavourable cardiovascular risk factors particularly in the case of PGMs. In contrast, with maternal grandparents, the association patterns are either U-shaped or inverse. LBW babies are associated significantly with stroke and diabetes in MGMs.
Some of the patterns of association, although not always statistically significant, fit with already published evidence, in that maternal grandparents’ relationships are with LBW or U-shaped patterns. This was also reported by Smith et al.Reference Smith, Wood, White, Pell and Hattie27 in relation to combined maternal grandparental CVD or mortality, by Manor and KoupilReference Manor and Koupil28 in relation to maternal grandparental circulatory and all-cause mortality and Naess et al.Reference Naess, Hoff and Lie30 in relation to MGMs’ cardiovascular status. Although McCarron et al.Reference McCarron, Davey Smith and Hattersley29 found a direct association between maternal grandparents’ diabetes status and grandchildren's birth weight, they also reported a U-shaped association between MGMs’ diabetes and mothers’ birth weight. This inverse association between mothers’ diabetes and children's birth weight has also been documented by many other parent–child studies.Reference Hypponen, Smith and Power7, Reference Davey Smith, Sterne, Tynelius and Rasmussen16
The MGM's association is through her daughter and is consistent with the intrauterine and the genetic hypothesis. Barker,Reference Barker39 the seminal author in this field, postulates that a 100-year period of nutritional flow from MGM and mother through the intrauterine route may influence the health for the lifetime of an individual. Moore and HaigReference Moore and Haig40 postulated that parents have different interests in the transfer of nutrients to their progeny, with maternally expressed genes targeting restriction of foetal growth. Manor and KoupilReference Manor and Koupil28 suggest that maternal genes may additionally influence the intrauterine environment and alter foetal growth. The twofold influence of MGM through intrauterine physiology and maternal genes may explain why her influence on chronic diseases such as stroke and diabetes is more prominent in this study. In contrast, the paternal interest is in an offspring that carries forth his genetic code, and thus paternally expressed genes influence foetal growth in the opposite direction.Reference Ong and Dunger41 This epigenetic influence may explain our paternal line results.
This cohort has previously published an association between MGMs’ BMI and infants’ birth weightReference Murrin, Segonds-Pichon and Fallon36 and also familial associations for children's BMI at age 5 years primarily along the three-generation maternal line;Reference Murrin, Kelly, Tremblay and Kelleher42 thus, the findings in our present analysis are also consistent with the cohort's previous findings.
An interesting finding in this study is the sex-specific association. A clear direct association is observed between the PGMs’ triglycerides levels and her granddaughters’ birth weight. The greater potential for triglycerides in predicting CVD and mortality in females than males is known.Reference Pilote, Dasgupta and Guru43 The observed association does suggest an X-mediated relationship as the granddaughter is 100% certain to have inherited at least one of her PGM's X-chromosomes, whereas the grandson inherits none. Although the X-chromosome contains only about 4.4% of human DNA, it contains about 8% of human genes and thus increases the likelihood of inheritance in PGM–granddaughter relationships. Moreover, as this particular X-chromosome undergoes relatively less crossing over, any carried epigenetic tag would be relatively conserved.Reference Fox, Sear and Beise44 The evidence on sex-specific trans-generational responses in humans comes from the Overkalix cohorts; Pembrey et al.Reference Pembrey, Bygren and Kaati32 showed that PGFs’ food supply was linked to the mortality-relative risk of grandsons only, whereas PGMs’ food supply was associated with mortality-relative risk of granddaughters only, and hypothesized a sex-chromosome-linked epigenetic transmission for paternal grandparents. They found no associations for maternal grandparents in their analysis.
The study has acknowledged limitations. The outcome events in Lifeways are as yet low and power considerations may well be influencing the patterns seen. There are also incomplete data at the different time points. It is a small cohort and only some grandparents participated in the cardiovascular risk factor assessment phase. We described at recruitment that this was due to resource constraints rather than active refusal by grandparents to participate.Reference Kelleher, Fallon and Fitzsimon33 Nonetheless, some self-selection bias may be at play. It is possible that some of the patterns seen, for example the relationship between PGFs’ mortality and high birth weight, relate somehow to self-selection by already unwell grandfathers in the first instance to the study. However, that should not affect the fact that the would-be born babies with such grandfathers are larger, and the relationship in fact grew stronger after adjustment for possible confounding factors, inclusive of smoking and social position. Although there is also some bias towards the higher social position in the families with participating grandparents, adjustment for education level of cohort members made no appreciable difference to the findings, as also observed by other similar studies.Reference Friedlander, Paltiel and Manor6, Reference Smith, Pell and Walsh10, Reference Friedlander, Manor and Paltiel12, Reference Smith, Wood, White, Pell and Hattie27–Reference McCarron, Davey Smith and Hattersley29
This familial analysis required testing of a number of comparisons to investigate the hypothesis regarding differences by lineage and gender. Parent-of-origin differences in such associations have been reported previously,Reference Davey Smith, Hypponen, Power and Lawlor15 and gender differences in cardiovascular diseases are well known.Reference Pilote, Dasgupta and Guru43 Investigation into sex-specific effects in long-term disease risk associated with birth weight has been recommended for an understanding of intergenerational programming and inheritance patterns.Reference Drake and Walker22 However, considering that these comparisons may result in some significant findings by chance, a conservative approach to interpretation was adopted to allow for chance or false discoveries. Inferences were based on consistent patterns observed across levels and time points of analyses that could meaningfully be explained together and not on P-value of any single test. Rational multiple comparisons have been accepted in observational studies, and it has been argued that overcorrection such as by the Bonferroni method might lead to type 2 error.Reference Rothman45, Reference Savitz and Olshan46
The study had the advantages of a priori hypothesis and appropriate prospective study design. Only recognizable patterns that satisfied the criteria of consistency within the study and with previously reported studies, temporality in sequence of events, biological plausibility and coherence with existing knowledge of intergenerational mechanisms of risk transmission were accepted as true discoveries.Reference Hill47
In conclusion, these data from the Lifeways cohort show intergenerational patterns of association that differ according to maternal and paternal grandparental lines, in keeping with the evidence published to date but with the added value that three different sets of prospective associations are observed. It is of considerable importance that other cohort studies seek to add biological and genetic information on grandparents in their data collection strategies to elucidate further these important cross-generation mechanisms.
Acknowledgements
The Health Research Board of Republic of Ireland has funded all sweeps of this study. The Health Promotion and Policy Unit of the Department of Health and Children contributed funding to the cardiovascular risk factor assessments of grandparents.
The Lifeways cross-generation cohort study is overseen by a multi-disciplinary steering group, whose members are (in alphabetical order), Professor Gerard Bury, Professor Leslie Daly, Professor Sean Daly, Dr Orla Doyle, Dr Una B. Fallon, Dr Frances B. Hannon, Dr Howard Johnson, Dr Lucy J. Jessop, Professor C. Cecily Kelleher, Professor B. Gerard Loftus, Professor John J. Morrison, Professor Andrew W. Murphy, Dr Celine Murrin, Dr Isabelle Niedhammer, Dr John O'Brien, Professor Helen Roche, Dr Aakash Shrivastava, Dr Mary Rose Sweeney, Professor Richard Tremblay and Dr Karien Viljoen. The participation of families is much appreciated.
Ethical approval for Lifeways study was obtained from ethical committees of Coombe University Hospital, Dublin, University College Dublin, Irish College of General Practitioners and University College Hospital Galway, Ireland.
AS and CK drafted the manuscript, with CM, JOB, KV, PH, TG providing critical revision. AS undertook data collection and analysis as a part of his PhD supervised by CK. CK, the principal investigator, designed the study and obtained funding. CM, JOB, KV, PH were involved in data collection and management. TG provided statistical inputs. All contributors have approved the final version of the manuscript.
