Genome-wide association study reveals a dynamic role of common genetic variation in infant and early childhood growth

Infant and childhood growth are dynamic processes characterized by drastic changes in fat mass and body mass index (BMI) at distinct developmental stages. To elucidate how genetic variation influences these processes, we performed the first genome-wide association study (GWAS) of BMI measurements at 12 time points from birth to eight years of age (9,286 children, 74,105 measurements) in the Norwegian Mother and Child Cohort Study (MoBa) with replication in 5,235 children (41,502 measurements). We identified five loci associated with BMI at distinct developmental stages with different patterns of association. Notably, we identified a novel transient effect in the leptin receptor (LEPR) locus, with no effect at birth, increasing effect on BMI in infancy, peaking at 6-12 months (rs2767486, P6m = 2.0 × 10−21, β6m = 0.16) and little effect after age five. A similar transient effect was found near the leptin gene (LEP), peaking at 1.5 years of age (rs10487505, P1.5y = 1.3 × 10−8, β1.5y = 0.079). Both signals are protein quantitative trait loci (pQTLs) for soluble LEPR and LEP in plasma in adults and independent from signals associated with other adult traits mapped to the respective genes, suggesting novel key roles of common variation in the leptin signaling pathway for healthy infant growth. Hence, our longitudinal analysis uncovers a complex and dynamic influence of common variation on BMI during infant and early childhood growth, dominated by the LEP-LEPR axis in infancy.


Infant and childhood growth are dynamic processes characterized by drastic changes in fat mass and body mass index (BMI) at distinct developmental stages. To elucidate how
genetic variation influences these processes, we performed the first genome-wide association study (GWAS) of BMI measurements at 12 time points from birth to eight years of age ( BMI patterns in infancy and childhood follow well-characterized trajectories: a rapid increase soon after birth until approximately nine months, the adiposity peak, followed by a gradual decline until approximately 4-6 years of age and then the adiposity rebound, when BMI starts to increase again until the end of puberty 1 . Recently, a study revealed that the most powerful predictor of obesity in adolescence is an increase in BMI between two and six years of age 2 , but the underlying cause for this remains unknown. To explore how common genetic variation influences these processes, we performed the first genome-wide association study (GWAS) of early growth in the population-based birth cohort, the Norwegian Mother and Child Cohort Study (MoBa) 3 (Supplementary Table 1). A total of 17,474 children in MoBa were genotyped in discovery and replication combined. The children's BMI was measured at birth, 6 weeks, 3, 6, 8 months, and 1, 1.5, 2, 3, 5, 7, and 8 years of age (Fig. 1a). We performed genotype quality control (QC), imputation using the Haplotype Reference Consortium (HRC), and phenotype QC, leaving 9,286 and 5,235 samples for the discovery and replication cohorts, respectively, all of Norwegian ancestry.
We conducted separate linear regression analyses of standardized BMI for each time point using an additive genetic model. The lead SNPs at independent loci reaching P < 10 -7 at one or more time points in the discovery sample were taken forward for replication (Table 1). This revealed a highly dynamic pattern of association during early growth. SNPs in five independent loci reached genome-wide significance, presenting peak association at different time points: (1) a novel, intronic SNP rs2767486 in the LEPR locus peaking at six months; (2) an intronic SNP, rs13035244, near ADCY3 peaking at one year; (3) an intronic SNP rs6842303 near LCORL peaking at 1.5 year; (4) an intergenic SNP rs10487505 near LEP peaking at 1.5 year; and (5) an intronic SNP rs9922708 near FTO peaking at seven years (Fig. 1b, c, and Supplementary   Fig. 1, 2).
The strongest association with BMI was found for rs2767486 at six months (P6m = 2.0 × 10 -21 , β6m = 0.16) in the LEPR/LEPROT locus. The locus strongly associated with BMI from three months of age, with effects peaking at 6-12 months, and waning from age three with little effect at eight years ( Fig. 1c, 2, and Supplementary Fig. 3a). We found no evidence of association at birth for rs2767486 or nearby markers in our data or in recent large publicly available GWASs of birth weight 4 and adult BMI 5,6 . Thus, this locus most likely affects BMI development primarily during infancy. Conditioning on rs2767486 revealed a putative additional signal in the LEPR locus, rs17127815 (P6m = 7.5 × 10 -5 after conditioning), that mirrored the association pattern of the main signal ( Supplementary Fig. 3b).
LEPR encodes the leptin receptor, which functions as a receptor for the adipose cell-specific hormone leptin. High leptin levels suppress hunger by interacting with the long form of the leptin receptor (OB-RL) in the hypothalamus 7 . The soluble form of leptin receptor (sOB-R), which is produced through ectodomain shedding of OB-RL in peripheral tissues, can bind leptin in circulation, and thereby reduce its effect on the central nervous system 8 . The LEPR locus has previously been implicated in monogenic morbid obesity 9,10 , severe childhood obesity 11 , age of menarche 12 , age of voice breaking 13 , levels of fibrinogen 14 and C-reactive protein 15 , several blood cell count traits 16,17 , and plasma sOB-R levels 17,18 . To test whether any of the established variants for these traits explains the observed association with BMI in infancy, we repeated the analysis conditioning on the top SNPs reported in these studies. The association with infant BMI remained unaffected by conditioning on these SNPs, except for rs2767485 ( Supplementary   Fig. 4a), the strongest pQTL for sOB-R-plasma levels in adults 18 . Intriguingly, this SNP is located only 12.2 kbp upstream our top SNP rs2767486, with strong LD (r 2 = 0.9) between the BMIraising and the sOB-R-increasing alleles.
The strong association between variants in the LEPR locus and infant BMI suggests an important role of leptin signaling in early growth. The genome-wide significant association with infant BMI for rs10487505 located 20 kbp upstream of LEP is therefore noteworthy. This SNP is a known pQTL for circulating leptin levels in adults 19 . The leptin-increasing allele from Kilpeläinen et al. 19 is associated with lower infant BMI in our data. The effect presents a riseand-fall pattern, rising during the 3-12 months period when the LEPR signal is at its plateau, reaching its peak at 1.5 years (P1.5y = 1.3 × 10 -8 , β1.5y = 0.08) before waning (Fig. 1c, Supplementary Fig. 5c, 6). Children homozygous for the alleles associating with higher sOB-R and lower leptin levels exhibited higher mean standardized BMI (+0.65) than children homozygous for the opposite alleles (Fig. 3).
We identified a novel association with BMI in the LCORL locus for rs6842303, presenting a similar rise-and-fall pattern with peak effect at 1.5 years (P1.5y = 7.5 × 10 -9 , β1.5y = 0.09) (Fig. 1c, Supplementary Fig. 5b, 6). Previously, this marker has been associated with related traits such as birth weight, birth length, infant length, and adult height. Interestingly, rs6842303 has also been associated with peak height velocity in infancy 20 , but no association was reported in the largest adult BMI GWASs to date 5,6 . This supports our finding of a transient effect of LCORL in early growth.
The second strongest signal was found at the ADCY3 locus. Biallelic mutations in ADCY3 have recently been found to cause severe syndromic obesity 21,22 . ADCY3 is known to interact with MC4R, and rare mutations in MC4R account for 3-5% of severe obesity 23 . The lead ADCY3 SNP, rs13035244, showed no association at birth, gradually became genome-wide significant with a peak effect between one and 1.5 years (P1y = 7.9 × 10 -13 , β1y = 0.10), and then stabilized during the course of childhood (Fig. 1c, Supplementary Fig. 5a, 6). This result is in agreement with a previous study of growth trajectories in children from one to 17 years of age 24 .
In contrast to the rise-and-fall pattern reported here for signals in the LEPR, ADCY3, LEP, and LCORL loci, the FTO risk allele displayed a trend towards a slightly negative effect around adiposity peak before gradually turning positive from three years of age, reaching genome-wide significance at seven years (P7y = 2.8 × 10 -12 , β7y = 0.12), in agreement with Sovio et al. 25 . (Fig. 1c, Previous studies have suggested a tight genetic overlap between child and adult BMI, but the details of this relationship across the first years of life remain elusive 24,26 . We used LD score regression 27 in LD Hub 28 to quantify the shared genetic contribution between BMI at each of the 12 time points and other traits (Fig. 4a,b and Supplementary Fig. 7). Our data show that although adult BMI and other adult obesity traits normally associated with poor metabolic control were positively correlated with childhood BMI from age 5-8 years, this correlation was much weaker below the age of three years. Notably, the genetic correlation with a range of nonanthropometric traits varied substantially at infant age ( Supplementary Fig. 8). Polygenic risk score analyses across all time points for markers associated with birth weight 4 , childhood BMI 26 , and adult BMI 5,6 revealed similar patterns (Fig. 4c). We also used LD score regression to estimate the SNP-based heritability of BMI measurements across infancy and childhood. The LD score regression-based heritability estimates varied with age, with relatively modest levels at birth and during the adiposity rebound, and high levels when adiposity is high, i.e. around adiposity peak and from seven years of age onwards (Fig. 4a). This finding is supported by twinstudies that also show high heritability estimates for BMI in infancy, lower levels around four years of age, followed by higher estimates in later childhood 29 . Collectively, these results further indicate that the genetic mechanisms underlying BMI change from infancy to adulthood.
To our knowledge, we report the first GWAS with dense measurements of BMI during the first year of life. The few GWASs published on BMI in infancy and childhood mainly involve children above five years of age, i.e. during adiposity rebound 24,26 . These studies point toward a strong genetic correlation for BMI around adiposity rebound and adulthood. Our results confirm a strong overlap of the genetics of BMI from five to eight years and adulthood, however, this association is much less pronounced during infancy. Infant weight and height have considerable heritable components 30 . Our results suggest that there are distinct molecular mechanisms that dynamically and specifically influence weight gain in infancy, partly acting through leptin signaling.
Leptin has an important role in fetal growth, and is positively correlated with birth weight 31 .
Leptin levels are high at birth and decrease quickly, whereas sOB-R levels are low at birth and increase rapidly during the first postnatal days 32 . This pattern is hypothesized to be an important mechanism for suppressing leptin-induced energy expenditure during the first neonatal days. The sOB-R level remains very high during the first two years of life and then declines 33 , mirroring the association of LEPR with infant BMI observed in our study (Fig. 1c). An effect of genetic variant(s) on the level of sOB-R in infancy is therefore a possible causal mechanism underlying the association with BMI. An interaction between the LEPR-and LEPassociated variants with increased BMI in individuals who carry both the sOB-R-raising and leptin-lowering alleles would further support a mechanism where sOB-R in circulation sequesters leptin, reducing its membrane receptor activation, hence promoting energy intake during infancy. The SNPs associated with increased BMI during infancy near LEPR and LEP are not known to affect adult BMI. In fact, they are not in LD with any marker associated with adult diseases, and might thus promote healthy weight gain during infancy, a notion further supported at the genome level by LD score regression. This result is further supported by a recent independent study 34 suggesting that SNPs in the LEPR/LEPROT locus are associated with BMI at the adiposity peak.

Study population
The

Pre-phasing and imputation
Prior to imputation, insertions and deletions were removed to make the dataset congruent with If for an individual of sex s, two consecutive height values, h i and h i +1 presented a decrease in height, i.e. h i +1 < h i , this was considered an artefact and corrected as follows.
If the individual presented three or more other height measurements, h j with j ≠ i and j ≠ i+1, for each j the corresponding height at i and i+1 was estimated by interpolating the height curve using the ratios as in equation 2: Where m s , i is the median and F s ,i −1 the empirical quantile function of the heights at i of individuals of sex s presenting at least three values before age two (exclusive) and at least two ot h erwise (6) Subsequently, height and weight missing values were imputed from the individual height and weight curves at all ages for individuals presenting at least three values before age two (exclusive) and at least two values after age two (inclusive), and until age two (exclusive), for individuals presenting at least three values before age two (exclusive). A missing value at i was

Statistical analyses
Genome-wide analyses were performed using SNPTEST v.2.5.2 using dosages of alternate allele with an additive linear model using sex, batch, and ten principal components as covariates. LD

Ethics
The study was approved by the Regional Committee for Medical and Health Research Ethics in Norway (#2012/67).       (ii) the genomic coordinates in build GRCh37; (iii) the nearest gene; (iv) the age at peak, i.e. lowest P-value; (v) the BMI-increasing and non-increasing alleles, EA and Non-EA, respectively; (vi) the BMI-increasing allele frequency (EAF); (vii) the regression beta ( ), standard error (SE), and β associated P-value for discovery, replication, and meta-analysis; and (viii) heterogeneity I 2 .