Differences in the Heritability of Growth and Growth Velocity During Infancy and Associations With FTO Variants

Abstract

While the associations of common variants in the FTO gene with obesity have been widely replicated in adults, there is conflicting evidence regarding their effects in infancy. We hypothesize that the genetic influences on growth traits vary during infancy, and that conflicting results may stem from variation in the ages at which FTO associations have been examined. Using longitudinal weight and length data at 0, 1, 3, 6, 9, 12, 18, 24, 30, and 36 months in 917 (444 females, 473 males) family members from the Fels Longitudinal Study, we used a variance components–based approach (SOLAR) to: (i) examine differences in heritability (gene-by-age interaction) in weight, length, relative weight (BMI and ponderal index (PI)) and instantaneous weight and length velocities over the course of infancy, and (ii) test whether a common FTO variant (rs9939609) was associated with infant growth at three ages (maximum trait heritability, birth and 36 months). All heritabilities at birth (of 39–74%) were significant (P < 3.9 × 10⁻¹⁰), but changed with age (gene-by-age interaction, P < 0.05). Weight, relative weight, and weight velocity reached maximum heritabilities (of 76–89%) at 6–9 months, while length and length velocity reached maximum heritabilities (of 96–99%) at 18–30 months. We found no association of rs9939609 with growth status or velocity measured at any age (P > 0.11). This study for the first time demonstrates the fluctuation of genetic influences on infant growth, but further work is required to determine which gene variants explain the strong additive genetic effects observed.

INTRODUCTION

There is extensive evidence that the risk of chronic diseases such as obesity, diabetes, coronary heart disease, and cancer is developmentally programmed. That is, risk for disease is affected by the interaction of both prenatal and early postnatal environmental exposures with the fetal and infant genome (1). Despite the importance of early life gene-by-environment interaction on health and disease, there is relatively little genetic epidemiologic research conducted in infancy and early childhood. For example, while evidence of genetic effects on birth weight is strong (2–4), documentation of the heritability and measured genotypic contributions to growth and other markers of health during the early postnatal period remains fragmentary, despite the known importance of postnatal weight gain to future obesity risk (5,6). Previous work by our-selves and others have shown a high heritability (h²) for weight and length during infancy (6–9), but the genetic architecture of obesity-related measures such as relative weight or weight velocity, which are frequently used as markers of infant nutritional status and obesity risk, have not been examined.

BMI tracks strongly from adolescence into adulthood, but less strongly from early childhood (10). One explanation is that different genetic variants contribute to obesity in early life and in adulthood. Polymorphisms in the FTO (fat mass and obesity-related) gene are, to date, the most widely replicated common obesity-related genetic variants in adults (2,3,11,12), but evidence is mixed on their associations in infancy. Some studies have found a relationship between FTO variants and infant weight and adiposity (13,14), while others have not (3,15–17). We hypothesize that lack of replication of FTO effects in infancy may relate to the rapidly changing genetic and maternal environmental influences on growth traits that may increase the likelihood of finding effects at some ages more than others.

Using one of the largest existing family-based longitudinal studies of infant and childhood growth, the aims of the study were to: (i) provide a comprehensive analysis of the heritability of infant weight, relative weight, length and growth velocities, and the changes in their heritability during infancy (i.e., gene-by- age interactions), and (ii) test the hypothesis that a common variant in FTO is significantly associated with infant growth or growth velocities, particularly at ages when trait heritabilities are high and theoretically most likely to reveal the influence of measured genetic effects.

METHODS AND PROCEDURES

Study sample and measurements

Study subjects included 917 white participants (444 females, 473 males) in the Fels Longitudinal Study, a study of child growth and development and later cardiovascular disease risks conducted in the Dayton, OH area of the United States (18). The 917 participants were distributed among 121 nuclear and extended families and in total comprised of 5,393 relative pairs, of which 1,036 pairs were of first-degree relatives, 1,031 pairs were of second-degree relatives, 1,128 were of third-degree relatives, and 2,198 pairs were of less closely related relatives.

For inclusion in this analysis, participants were required to have at least three serial weight and length measurements taken between birth and age 42 months. Weight and recumbent length were measured by trained anthropometrists at target ages of 0, 1, 3, 6, 12, 18, 24, 30, 36, and 42 months. For measurements over 24 months of age, standing height was used in lieu of recumbent length.

Relative age of the infant was calculated as the age at measurement minus the exact target age and was included in all analyses as a covariate to account for small individual variations in the exact timing of the measurements, which varied no more than 2 weeks before or after the target age at birth and 1 month, no more than 6 weeks at 3, 6, 9, and 12 months, and no more than 3 months at 18, 24, 30, and 36 months. Gestational age, birth order, and maternal age were obtained from questionnaires administered by study staff to the mothers at their infant’s first exam. Means and ranges of the covariates used for analyses are listed in Table 1. Gestational age information was missing for ~30% of the subjects, and was imputed five times (19) using PROC MI in SAS (Cary, NC). The imputation model contained all measurements of weight, length, ponderal index (PI), sex, birth year, age at measurement, birth order, maternal age, and interactions of birth year by gestational age, maternal age, and sex, in order to conform to the potential structural complexity of the data. Heritabilities and s.e. of the heritabilities differed minimally across the five gestational age imputation runs. As the increases in s.e. estimates after adjusting for multiple imputations were negligible relative to the magnitude of the h² estimates, and using just one imputation did not influence the interpretation of results, we randomly chose one imputation set for all (univariate, gene-by-age and measured genotype) analyses.

Table 1.

Raw means, sample size, and ranges of covariates for study participants

	N	Mean	Minimum	Maximum
Sex (females)a	917		444 (48%)
Birth orderb	903	2	1	10
Year of birth (year)	917	1962.9	1928	2009
Gestational age (weeks)	670	39.6	27.2	47.1
Maternal age (years)	911	28.3	15.1	46.2

^aNumber (%) listed under mean;

^bMedian listed under mean.

A subset of 534 participants with available DNA was genotyped for single-nucleotide polymorphisms (SNPs) in the FTO gene (rs1421085, rs17817449, rs8050136, and rs9939609) using TaqMan assays (Applied Biosystems, Foster City, CA). These common variants were selected based upon their frequently observed association with adiposity traits in adults of European ancestry. At the phenotypic level, the 534 individuals did not have significantly different weights and lengths than the full sample, although on average, they were slightly older when measured at their visit at birth and one month (P < 0.02). Participants were also no different at the additive genetic level as tested in a gene-by-genotyped analysis. That is, the genetic variances of all weights and lengths for participants with SNP data were not significantly different from those without SNP data. All protocols were approved by the institutional review board of Wright State University and informed consent of the parents was obtained at each study visit.

Model fitting

Missing growth data (<5% of subjects at any given time point) occurred randomly due to missed appointments to the study center. That is, while one participant may be missing their 3-month visit, a different participant may have missed their 6-month visit. We modeled the serial weight and length data for all children in a mixed effects regression model using a fourth-degree polynomial function of age, implemented in SAS PROC MIXED, which was similar to a technique used by van Dommelen et al. (20) to analyze longitudinal growth data in the Dutch Twin Study. Allowing each curve parameter to be estimated simultaneously in a mixed effects regression model ensures a consistent sample size for each time point (age) and a reduction in error compared to the use of individual regression models (20). All age terms were included as both fixed and random effects, and models also included sex, birth year, birth year², age-by-sex and age-by-birth year interactions as fixed effects. Age³ and age⁴ random effects were not significant for length and were dropped out of the models for length. Residual and influence diagnostics indicated that model assumptions were not violated and estimates of fixed effects were not overly influenced by any single individual. Additionally, correlations between observed and predicted values for all growth metrics and at all ages were high (r > 0.88) indicating that these models accurately fit the observed patterns of individual infant growth.

Using the estimates for fixed effects and predictions for random effects from these models, predicted weight and length were obtained at each specific target age (e.g., birth, 1 month, etc.), and then used to calculate predicted BMI (weight (kg)/length (m²) and predicted PI (weight (kg)/ length (m³)) at each target age. We calculated the first derivative from each individual’s growth curve equation for weight and length to compute individual instantaneous velocities at each age except at beginning and end points (birth and 42 months). We computed velocities at age one day in lieu of at birth. The predicted static measurements and instantaneous velocities (Table 2) were then used in the following genetic analyses described below. Age 42 months was not included in the final analysis as it was predicted with less accuracy due to it being at the tail of the growth curve and there were fewer measurements available.

Table 2.

Predicted means (± s.d.) from the mixed modelling procedures for weight, length, relative weight, and for instantaneous velocities from ages 0–36 months

Age (months)	Static measures				Instantaneous velocities
	Weight (kg)a	Length (cm)a	PI (kg/m3 )	BMI (kg/m2)	Weight (kg/year)b	Length (cm/year)b
	n = 879	n = 891	n = 853	n = 853	n = 879	n = 891
0	3.4 ± 0.5	51.1 ± 2.1	25.2 ± 2.4	12.9 ± 1.2	11.0 ± 2.0	39.1 ± 3.4
1	4.2 ± 0.5	54.2 ± 2.1	26.6 ± 2.0	14.4 ± 1.1	10.0 ± 1.7	39.1 ± 3.4
3	5.8 ± 0.7	59.7 ± 2.3	27.0 ± 1.9	16.1 ± 1.1	8.2 ± 1.3	35.9 ± 3.0
6	7.5 ± 0.8	66.4 ± 2.3	25.8 ± 1.8	17.1 ± 1.2	6.1 ± 0.9	30.2 ± 2.5
9	8.8 ± 0.9	71.5 ± 2.4	24.1 ± 1.7	17.3 ± 1.2	4.5 ± 0.8	23.3 ± 1.8
12	9.8 ± 1.0	75.7 ± 2.5	22.7 ± 1.5	17.1 ± 1.1	3.3 ± 0.7	18.2 ± 1.5
18	11.1 ± 1.2	81.8 ± 2.8	20.3 ± 1.4	16.6 ± 1.1	2.2 ± 0.6	14.6 ± 1.3
24	12.2 ± 1.3	86.9 ± 3.1	18.5 ± 1.4	16.1 ± 1.1	2.1 ± 0.5	10.8 ± 1.1
30	13.3 ± 1.4	91.9 ± 3.3	17.1 ± 1.2	15.7 ± 1.1	2.3 ± 0.5	9.9 ± 0.8
36	14.4 ± 1.5	96.6 ± 3.5	16.0 ± 1.1	15.4 ± 1.0	2.2 ± 0.5	9.8 ± 0.7

PI, ponderal index (kg/m³).

^aMixed models adjusted for a random 4th degree age polynomial (centered at 1.5 years), and fixed effects of quartic age, sex, birth year (centered at 1960), birth year², age-by-sex and age-by-birth year interactions. Age³ and age⁴ random effects were not significant for length and therefore dropped out of the models for length.

^bMeasurement at 0 months estimated at 1 day.

Quantitative genetic analysis

Growth traits and velocities were inverse normalized where necessary to handle residual kurtosis. Maximum-likelihood variance components methods in SOLAR (21) were used to estimate narrow-sense heritabilities (h²) of each infant metric at ages 0, 1, 3, 6, 9, 12, 18, 24, 30, and 36 months adjusted for gestational age, sex, birth order, maternal age, birth year, and relative age. The age at maximum heritability was noted for each trait. Narrow-sense heritability, defined as the proportion of the phenotypic variance attributable to additive genetic effects, was estimated as h² = σ²_G/σ²_P, where σ²_G is the additive genetic variance and σ²_P is the total phenotypic variance.

Likelihood ratio tests were used to test if h² estimates were significantly different from zero. Gene-by-age interactions were also evaluated using likelihood ratio tests. Gene-by-age interaction effects occur when the genetic correlation (ρ_G), a measure of shared genetic influences (pleiotropy) between a trait measured at two different ages is significantly different from one, or when the genetic variances (σ_G) of a trait measured at two different ages is significantly different at the two ages (22). Trivariate quantitative genetic analysis was used to simultaneously estimate the genetic correlations and variances at three time points: birth, age at maximum trait h², and at 36 months of age in order to minimize the number of statistical tests performed. In SOLAR, for a trivariate analysis, the covariance is given by a 3 × 3 covariance matrix with elements defined as Ω_ab = 2Φρ_Gσ_gaσ_gb + Iρ_Eσ_eaσ_eb, where 2Φ is the kinship matrix that structures σ_ga and σ_gb, the variance due to the additive effects of genes for time points a and b; I is an identity matrix that structures σ_ea and σ_eb, the variance due to unmeasured, non-genetic factors in a and b; ρ_G is the additive genetic correlation between time points a and b, and ρ_E is the unmeasured environmental correlation between the two time points.

Specifically, to test for gene-by-age interaction effects or continuity of genetic control, we tested whether ρ_G estimates for each trait measured at the three different ages were significantly different from one (i.e., complete pleiotropy). For each trait measured at different times (i.e., ages), complete pleiotropy occurs when the same gene (or set of genes) that influences a trait at one age also influences the trait at other ages. In this gene-by-age context, we tested for common genetic effects for each trait at the three selected age points. Thus, if ρ_G is different from one it is concluded that at least some unique genes are influencing the same trait at different ages, and a gene-by-age interaction effect exists. In situations of no pleiotropy, genetic correlations are not significantly different from zero, and the genes that influence a trait at one age are wholly different from the genes that influence the trait at the second age. Incomplete pleiotropy occurs when the genetic correlation is different from both zero and one. In this case, some, but not all, of the genes that influence a trait at one age, also influence the trait at other ages.

Similarly, differences in h² by age were tested by examining differences in the decomposed variances of each trait at each age. We tested whether genetic variances (σ_G) or environmental variances (σ_E) were significantly different at the different ages. By modeling parameters of σ_G and σ_E, it could be determined if differences in h² were due to changes in the effect of polygenes (σ_G), residual environmental effects (σ_E), or both. Differences in the genetic variances estimated at different ages can be interpreted as an indicator of differences in the expression of polygenes and are therefore indicative of gene-by-age interaction effects.

Measured genotype approach

Associations between FTO variants and infant growth metrics measured at birth, age at maximum h², and age 36 months were tested using the measured genotype approach using a maximum-likelihood, variance components–based, procedure implemented in SOLAR (21) that accounts for the nonindependence among family members and assumes additivity of allelic effects (21,23). Additive genetic effects and residual environmental effects were modeled as random effects, while sex, gestational age, maternal age, birth order and the FTO SNPs were modeled as fixed effects (21,23). The quantitative transmission dis-equilibrium test (24) was used to examine potential hidden population stratification.

RESULTS

Heritability of infant growth traits

The heritability of infant growth metrics was significant (P < 3.9 × 10⁻¹⁰) and additive genetic effects explained from 39% of covariate-adjusted variance for PI at birth (h² = 0.39, s.e. of h² = 0.08) to 74% of the phenotypic variance for length at birth (Figure 1 and see Supplementary Table S1 online). Heritability estimates of all traits increased during early infancy to mid-infancy, explaining 75–92% of the phenotypic variance by 1 year. Maximum h² occurred at age 9 months for weight, age 30 months for recumbent length, and age 6 months for PI and BMI (Figure 1).

Figure 1.

Heritability estimates for weight, length, BMI, and PI at ages 0, 1, 3, 6, 9, 12, 18, 24, 30, and 36 months in the Fels Longitudinal Study. Age at maximum heritability was 9 months for weight, 30 months for length, 6 months for PI, and 6 months for BMI. Heritability estimates can also be found in Supplementary Table S1 online. PI, ponderal index (kg/m³).

Heritability estimates of the instantaneous velocities for weight and length are shown in Figure 2 (and see Supplementary Table S1 online). Heritabilities of weight velocity were 0.64 at birth (i.e., one day of age), 0.78 at 6 months (maximum h²), and 0.58 at 36 months (all P < 3.0 × 10⁻¹²). Heritability of length velocities were 0.60 at birth (i.e., one day of age), 0.99 at 18 months of age, and 0.52 at 36 months (all P < 2.2 × 10⁻¹³).

Figure 2.

Heritability estimates for instantaneous weight and length velocities at ages 0, 1, 3, 6, 9, 12, 18, 24, 30, and 36 months in the Fels Longitudinal Study. Age at maximum heritability is at 6 months for weight velocity and 18 months for length velocity. Instantaneous velocity at age 0 estimated at one day. Heritability estimates can also be found in Supplementary Table S1 online.

Cross-trait bivariate quantitative genetic analyses (see Supplementary Table S2 online) were conducted to determine if each trait was influenced by a set of unique (non-pleiotropic) or shared (pleiotropic) genes at each age. While little evidence for genetic pleiotropy existed between length and BMI at each age, incomplete pleiotropy existed between all of the other traits, demonstrating that, as expected, the proportion of shared genes is high among infant weight, BMI, and PI.

Gene-by-age interaction

Differences in heritability of the traits at different ages were formally tested using trivariate quantitative genetic analyses for each trait at birth, age at maximum trait h², and age 36 months, and show a dynamic picture of genetic and environmental influences on growth during the infancy period (Table 3). The genetic correlations (ρ_G) among all traits at birth, maximum h², and age 36 months were significantly greater than zero (all P < 0.01) and significantly less than 1.0 (all P < 0.000004), indicating that some unique and some shared genes control growth traits and growth velocities over time (incomplete pleiotropy). Thus, gene-by-age interaction effects are present for each growth trait and velocity examined.

Table 3.

Gene-by-age interactions in weight, length, relative weight, and instantaneous velocities of weight and length at birth, maximum h² and 36 months of age

	Weight (kg)a	Length (cm)a	PI (kg/m3)a	BMI (kg/m2)a	Weight velocity (kg/year)a,b	Length velocity (cm/year)a,b
ρGc ± s.e.
Birth—maximumd h2	0.59 ± 0.07	0.66 ± 0.06	0.58 ± 0.09	0.49 ± 0.10	0.76 ± 0.05	0.69 ± 0.07
Birth—36 months	0.46 ± 0.07	0.65 ± 0.06	0.51 ± 0.11	0.39 ± 0.10	0.52 ± 0.08	0.32 ± 0.12
Maximum— 36 months	0.83 ± 0.03	1.00 ± 0.00	0.69 ± 0.07	0.66 ± 0.06	0.55 ± 0.08	0.73 ± 0.05
σG ± s.e.
At birth	0.68 ± 0.05e	0.65 ± 0.05e	0.63 ± 0.07e	0.62 ± 0.06e	0.73 ± 0.06e	0.51 ± 0.05e
At maximumd h2	0.80 ± 0.04f	0.93 ± 0.03f	0.85 ± 0.07f	0.85 ± 0.06f	0.79 ± 0.05e	0.93 ± 0.03f
At 36 months	0.85 ± 0.04f	0.93 ± 0.03f	0.79 ± 0.06e,f	0.78 ± 0.05f	0.74 ± 0.06e	0.72 ± 0.05g
σE ± s.e.
At birth	0.49 ± 0.05e	0.52 ± 0.05e	0.73 ± 0.05e	0.67 ± 0.05e	0.57 ± 0.06e	0.43 ± 0.05e
At maximumd h2	0.41 ± 0.06e	0.26 ± 0.06f	0.49 ± 0.08f	0.48 ± 0.07f	0.44 ± 0.06f	0.12 ± 0.07f
At 36 months	0.43 ± 0.05e	0.26 ± 0.06f	0.59 ± 0.06e,f	0.58 ± 0.05e,f	0.64 ± 0.05e	0.60 ± 0.04g

PI, ponderal index (kg/m³).

^aVariables inverse normalized and adjusted for sex of infant, relative age of infant at measurement, gestational age of infant at birth, birth order of infant, maternal age at birth, and birth year of infant.

^bMeasurement for birth estimated at 1 day.

^cAll ρ_G are significantly different from 0 and 1 (P < 0.01). Genetic correlations for a trait measured at two different ages that are significantly less than one (ρ_G < 1) are indicative of gene-by-age interaction, that is, some unique polygenes influence the trait at one age but not at the other age.

^dAge at maximum heritability was 9 months for weight, 30 months for length, 6 months for PI, 6 months for BMI, 6 months for weight velocity, and 18 months for length velocity.

^e,f,gStandard deviations or variances (σ_G, σ_E) within each trait with different superscripts are significantly (P < 0.05) different from each other. Unequal variances are indicative of gene-by-age interaction.

Tests for differences in the genetic variance (σ_G) and the residual environmental variance (σ_E), at each age indicated that the changes in heritability from birth to maximum h² were generally due to both significant increases in σ_G and significant decreases in σ_E. In the case of weight and weight velocity, however, only one of these components significantly changed (see Table 3). From age at maximum h² to 36 months, σ_G and σ_E were relatively stable for weight, length, PI and BMI, while σ_E increased for weight and length velocity. The σ_G decreased for length velocity over the same period.

Association of FTO variants with infant growth traits

The four genotyped FTO SNPs were in Hardy–Weinberg equilibrium (P > 0.66). Given that they exhibited high linkage dis-equilibrium (r² > 0.96 for all), we tested only the association with the rs9939609 SNP (T > A), the SNP most often reported in the literature as being associated with BMI. The minor allele frequency of the A allele was 39.9%. Trait means by SNP genotype, as well as the significance of the SNP for each metric at birth, maximum h² and 36 months are presented in Table 4. There was no evidence of population stratification (P > 0.13). No association of rs9939609 with infant weight, length, relative weight, or instantaneous velocities (Table 4) was found at any time point (all P > 0.11). The maximum variance explained by the SNP marker was h²_m = 0.009 (0.9%) for length velocity at 18 months.

Table 4.

no association of rs9939609 Fto snP with early growth in 534 Fels longitudinal study subjects

	Weight (kg)a	Length (cm)a	PI (kg/m3)a	BMI (kg/m2)a	Weight velocity (kg/year)a,b	Length velocity (cm/year)a,b
Birth
h2m	0.000	0.001	0.000	0.0004	0.000	0.001
μTT (± s.e.)	3.48 ± 0.03	51.73 ± 0.16	25.12 ± 0.20	12.98 ± 0.09	11.68 ± 0.17	40.50 ± 0.18
μTA (± s.e.)	3.46 ± 0.03	51.64 ± 0.13	25.09 ± 0.16	12.94 ± 0.08	11.67 ± 0.14	40.45 ± 0.15
μAA (± s.e.)	3.44 ± 0.04	51.56 ± 0.20	25.06 ± 0.25	12.91 ± 0.12	11.67 ± 0.21	40.39 ± 0.23
P valuec	0.49	0.50	0.86	0.66	0.97	0.70
Maximumd
h2m	0.002	0.004	0.0000	0.0000	0.006	0.009
μTT (± s.e.)	9.26 ± 0.08	92.67 ± 0.29	25.68 ± 0.17	17.31 ± 0.11	6.38 ± 0.08	10.54 ± 0.08
μTA (± s.e.)	9.19 ± 0.07	92.42 ± 0.24	25.65 ± 0.14	17.26 ± 0.09	6.29 ± 0.06	10.45 ± 0.07
μAA (± s.e.)	9.11 ± 0.10	92.18 ± 0.35	25.62 ± 0.20	17.21 ± 0.13	6.20 ± 0.09	10.36 ± 0.10
P valuec	0.21	0.24	0.81	0.52	0.11	0.14
36 months
h2m	0.004	0.004	0.0003	0.0003	0.0003	0.002
μTT (± s.e.)	14.83 ± 0.14	97.38 ± 0.30	15.99 ± 0.10	15.57 ± 0.10	2.20 ± 0.05	8.57 ± 0.06
μTA (± s.e.)	14.68 ± 0.11	97.11 ± 0.25	16.03 ± 0.08	15.55 ± 0.08	2.21 ± 0.04	8.53 ± 0.05
μAA (± s.e.)	14.54 ± 0.17	96.84 ± 0.37	16.08 ± 0.12	15.53 ± 0.12	2.23 ± 0.06	8.49 ± 0.07
P valuec	0.16	0.23	0.55	0.75	0.66	0.35

h²_m, SNP specific heritability; PI, ponderal index (kg/m³); SNP, single-nucleotide polymorphism; μ_TT, μ_TA, μ_AA, means of TT, TA and AA SNP genotypes ± s.e., respectively.

^aVariables inverse normalized and adjusted for sex of infant, relative age of infant at measurement, gestational age of infant at birth, birth order of infant, maternal age at birth, and birth year of infant.

^bMeasurement for birth estimated at one day.

^cP value is for h²_m.

^dAge at maximum heritability was 9 months for weight, 30 months for length, 6 months for PI, 6 months for BMI, 6 months for weight velocity, and 18 months for length velocity.

DISCUSSION

Despite a strong and growing interest in understanding the interaction of genetic and prenatal influences on childhood obesity, as seen in the design of the upcoming National Children’s Study of 100,000 children born in the United States (25), understanding of the genetic contributions to commonly collected infant growth and health traits remains fragmentary. This information is useful for guiding investigations into the genetic and epigenetic influences on early obesity-related traits, as it may help to identify infant growth traits that have the greatest likelihood of revealing genetic effects and may also be used to target nutritional interventions to periods of growth when environmental factors play a greater role.

Our estimates of total additive h² for birth weight and birth length are strong, and, like other family studies (8,9), our estimates are higher than those reported from twin studies (see review in ref. 2). There are a number of reasons why heritability estimates may differ by study design, but in the case of twin studies, the lower h² of birth weight and length may reflect greater prenatal environmental stressors experienced by twins. Not only are twins frequently born preterm, they are also more frequently growth-restricted relative to their gestational age at birth than singleton infants, stemming from some degree of placental insufficiency in multiple pregnancies, which may affect the twins unequally (26). Although twin studies provide uniquely powerful evidence regarding the role of shared maternal and postnatal environment vs. shared genetic influences, early growth traits in twins are nonetheless markedly different from singletons. Longitudinal extended pedigree studies such as the Fels Longitudinal Study provide an important perspective for infant (singleton) growth.

Wardle et al. (27) noted high heritability estimates for BMI and waist circumference in children born during the obesity epidemic that were similar to those from much earlier studies. Significant secular trends in childhood BMI have been previously reported in the Fels Longitudinal Study (28), however, the present data indicate that even after accounting for birth year, infant weight, BMI, and PI, are nonetheless highly heritable. It is sometimes assumed that because average weight and BMI have increased greatly during the past 30 years due to changes in the food and activity environment, genetic effects are relatively unimportant in explaining the risk of current human variation in obesity and its related traits. However, the present study, which followed families from 1929 to the present, clearly shows that even after accounting for secular trends towards increasing body size, the contribution of genetic factors to the individual differences in growth is very high.

As relative weight is rarely examined in studies of the genetics of infant growth, there are few studies with which to compare the heritability of relative weight (BMI and PI) against weight and length. Beardsall et al. (29) found a h² of 0.72 for BMI at birth, which was higher than their estimate for weight (h² = 0.43) at birth. The only h² reported for PI to date was very low (h² = 0.06), and not significantly different from zero (30). To our knowledge, only one other study has examined weight and length velocities (20). In this twin study, heritability estimates of instantaneous weight (h² = 0.57–0.63) and length (h² = 0.36–0.44) growth velocities at 1 year of age were lower than our estimates. Our study is the first to report growth velocities at different ages for pedigreed data.

In our study sample, there was a significant increase in the heritability of growth traits during the first postnatal year, which was due to simultaneously increasing genetic variance (σ_G) and decreasing environmental variance (σ_E). Some indications of increases in the heritability of infant weight and relative weight have been previously reported (20,30). Together, these studies and our data point to the timing of the “canalization of development” (31), or when the genetically based developmental course is most strongly manifest. It has long been held that genetically based growth trajectories do not become completely established in humans until after 2 years of age (32). The increasing σ_G in our data suggests that canalization is established sooner than that; canalization appears to be established by 6–9 months for weight and relative weight, and by 30 months of age for length. Indeed, by 18 months, the heritabilities for length and length velocity are remarkably high, and close to one. It is important to remember, however, that heritabilities are estimates with associated errors and are population specific. The Fels Longitudinal Study population is generally well nourished and free from significant disease and therefore our results may demonstrate the upper limits of such heritability estimates.

Previous work suggests that rapid growth in early infancy is more deleterious for future obesity risk (33) and insulin resistance (34) than is rapid growth in later infancy. The rapidly increasing genetic variance for infant weight, length, and their respective velocities that we observed over the first postnatal months suggests why weight and relative weight might be more responsive to environmental perturbation (e.g., formula feeding, dietary intervention effects) earlier than later in infancy.

FTO is predominantly expressed in the hypothalamus and thought to play a role in energy homeostasis (35,36). FTO variants have been found to predict child appetite, satiety, and food intake behavior (37), as well as BMI and adiposity in childhood (3,15–17,38). These studies suggest that FTO does not exert its influence on obesity until later in childhood, somewhere between 6 and 11 years of age (3,15–17,38). Haworth et al. (38) found an increasing influence of FTO on BMI as heritability increased, and we similarly hypothesized that the disputed effect of FTO on weight status could be explained by focusing on specific periods during infancy and early childhood when heritability was highest. However, we found no influence of rs9939609 on weight, length, relative weight, or instantaneous velocities in weight or length at any age during infancy. Similarly, some other studies also have been unable to identify a significant effect of FTO at this young age (15,16,38). For instance, in a large cohort (>20,000) of European American children 2–18 years of age, none of the FTO variants that were associated with BMI s.d. score at ages 6–18 were significant in children at 2–5 years of age (16). However, in a study of 225 Spanish infants, those carrying the T allele at rs9939609 had higher weight, PI, total fat, abdominal fat, and visfatin levels by 2 weeks of age (13), and Gemma et al. (14) found that AA homozygotes for the rs9930506 FTO SNP had lower weights at birth compared to infants who carried the G allele. The latter study was conducted in a very small sample (total n = 82), however, that included large numbers of both small and large for gestational age infants (14). Additional work is required to understand how FTO expression relates to the FTO genotype and energy metabolism in different tissues and at different developmental periods in humans (39).

More generally, a comprehensive examination of the genetic, intra-uterine, and postnatal environmental determinants of infant growth is critical, given its relationship to later obesity and cardiometabolic risk (5). Recent evidence from the EDEN study (40) showed that paternal BMI effects on infant growth velocity increase in the first 3 months, while maternal BMI effects were only important at birth. Genes involved in growth are often imprinted, and there is some evidence that paternally expressed genes (such as IGF2) may be more strongly expressed postnatally. Our present study did not examine maternal and paternal effects on prenatal and postnatal growth, but it is possible that part of the increasing genetic variance in infant size may reflect the postnatal manifestation of paternally expressed growth regulation, or other epigenetic changes in gene expression occurring dynamically over the period of infancy, as has been suggested recently (40).

Limitations of the study reported here are that the results cannot be generalized to nonwhite populations, or to the current US white population of infants because the average birth year in our study was 1962. However, our quantitative genetic results are consistent with the findings of more recently born cohorts of Hispanic (8) and UK white European children (27). Although our study is among the largest of family studies of infant growth in singletons, we were somewhat underpowered to detect FTO effects. Given the pedigree structure of the 534 individuals with SNP data, we had ~60% power to detect a SNP explaining 1.0% of the variance of in a trait with an additive heritability of 0.73 (the approximate mean of all heritabilities listed in Supplementary Table S1 online). Nonetheless, our findings agree with those of much larger studies that have failed to find significant associations between obesity-related traits and FTO variants before age 5 (15,16,38).

In summary, we found that the genetic and residual environmental influences on infant growth significantly change from birth to 3 years, with maximum heritability occurring at age 6–9 months for weight, relative weight, and weight velocity, and at 18 and 30 months for length and length velocity. Common FTO variants do not account for a significant proportion of this heritability. Due to the lower heritability estimates of weight, relative weight, and weight velocity at birth, environmental interventions to reduce rapid infant weight gain should theoretically have their greatest potential for effect before the age of 6–9 months, when heritabilities become high.