Effect of maternal nutrient intake during 31–37 weeks gestation on offspring body composition in Samoa

Abstract

Background:

Pregnancy dietary intake may be associated with newborn body composition, a predictor of future obesity. In Samoa, an energy-dense diet contributes to an alarming prevalence of adult obesity. Identifying associations between pregnancy nutrition and infant body composition in this setting may guide strategies to mitigate intergenerational transmission of obesity risk.

Aim:

To examine dietary macro and micronutrient intake of Samoan women during the third trimester of pregnancy and associations with infant body composition.

Subjects and methods:

At 34–41 weeks of gestation, we measured dietary intake from the prior month using a Food Frequency Questionnaire (FFQ). Dual-energy X-ray absorptiometry (DXA) measured infant body composition at 1–14 days. We used multivariable linear regression models accounting for confounders to identify independent effects of nutrient intake on infant body composition.

Results:

After adjusting for maternal body mass index, age, gravidity, infant age, and sex, a respective 0.2g increase and 0.2g decrease in infant bone mass was associated with fibre and saturated fat intake. Increased protein intake was associated with 0.02g decrease in bone mass.

Conclusions:

While maternal dietary intake was not associated with infant adiposity or lean mass, we observed an effect on bone mass whose role in regulating metabolic health is overlooked.

Introduction

Obesity is a growing global epidemic, and Polynesians are at particularly high risk relative to other populations: average body mass index (BMI) is 32.2 kg/m² for women and 29.2 kg/m² for men, compared to global averages of 24.2 kg/m² and 24.4 kg/m², respectively (NCD Risk Factor Collaboration, 2014). The high prevalence of obesity in Samoa, a Pacific Island nation where >80% of adults are either overweight or obese (Hawley, Minster, et al., 2014), has been attributed to changes in the Samoan nutritional environment in the last 50 years with a decline in subsistence agriculture and concurrent increases in imported foods leading to energy-dense, micronutrient-poor diets (Galanis, McGarvey, Quested, Sio, & Afele-Fa’Amuli, 1999; Hawley & McGarvey, 2015).

This is a concern from a developmental origins of health and disease perspective, which suggests that developing perinatal infant biology is highly sensitive to environmental stimuli and metabolic traits are programmed for life (Gluckman & Hanson, 2006; Kuzawa & Quinn, 2009). One major factor proposed to influence fetal programming is the nutritional status of the mother. Maternal BMI (a proxy for percent body fat) before and during pregnancy has been clearly linked to infant adiposity measured via tissue scans during perinatal development (Hull et al., 2008; Hull, et al., 2014; Sewell et al., 2006). More recently, as a next step, researchers have become interested in whether diet during pregnancy influences infant body composition because of its effect on maternal nutritional status and its potential for modification in behavioural health interventions. Nutrient intake during late-gestation has been proposed to have the greatest effect on infant body composition (Brei, et al., 2018), perhaps, in part, because 90% of fetal fat deposition occurs during the last ten weeks of pregnancy (Poissonnet, Burdi, & Garn, 1984; Widdowson, 1968).

While interest has grown there is still little consensus on the effect maternal diet at the macro- or micronutrient level on infant body composition. For example, Brei et al. (2018) measured macronutrient intake using 7-day dietary records during the 32^nd week of gestation in 167 women from Munich, Germany, and found a negative association between maternal dietary fat consumption and fat mass at birth. Conversely, Crume et al. (2016) and Renault et al. (2015) found positive associations between fat intake and fat mass at birth in their respective samples. Crume et al. measured dietary intake using a 24-hour dietary recall administered at least once between 8–32 weeks gestation in 1040 women from Aurora, CO, USA, and Renault et al. used a FFQ administered at least once at either 11–14 or 36–37 weeks to 222 obese pregnant women from Copenhagen, Denmark. The inconsistent results across these studies could be related to technical error limitations in measuring infant body composition using anthropometric techniques including weight, length and skinfold thicknesses (Ulijaszek & Kerr, 1999). They could also be attributed to the heterogeneity of samples, which vary in ethnicity and maternal adiposity.

This study addresses limitations of previous studies by using Dual-Energy X-ray Absorptiometry (DXA) to measure compartmentalised early infant body composition, providing ‘gold standard’ estimates of fat and fat-free (bone and lean) mass in seeking to detect associations with maternal dietary intake in the third trimester of pregnancy. In addition, our sample includes participants who had four Samoan grandparents—a level of ethnic homogeneity that is difficult to attain in other nations, which reduces influence from genetic or population-level factors that may affect infant body composition (Wells, 2014). Further, while the maternal sample is higher in adiposity compared to others, we acknowledge this prominently so that future use of our findings will take this into consideration. Overall, our study seeks to illuminate how an obesogenic environment, similar to alterations in nutritional ecologies in many other low/middle income countries, may have direct impacts on infant body composition as early as during gestation.

Subjects and methods

Setting

Samoa is a Pacific Island nation placed in the World Bank’s upper-middle income category in 2019, with a Gross Domestic Product (GDP) of $6089.30 USD (Samoa GDP Per Capita PPP, 2018). The island of Upolu, where 75% of the population lives and all participants lived, is divided into three census regions with varying degrees of urbanicity and exposure to the nutrition transition, including the Apia Urban Area (AUA; most exposed), sub-urban North West Upolu (NWU; variably exposed), and rural Rest of Upolu (ROU; least exposed). Apia is the capital and largest city on the island, where around 19% of the population live. Study participants were recruited from the Tupua Tamasese Meaole (TTM) Hospital in Apia, which is the only tertiary hospital in Samoa and the major referral centre on Upolu.

Study design, population and sample

This study utilises data from a convenience sample of n=107 Samoan mother-infant dyads participating in a prospective birth cohort study: “Foafoaga O le Ola” (the “Beginning of Life” study). The main aim of the study was to assess how maternal diet during the last trimester, breastfeeding duration, and presence of a CREBRF gene variant (Minster et al., 2016) act independently and interact to influence adipose tissue growth in infants from birth to 4 months (since adiposity has been shown to peak at around this time in Samoan infants (Hawley et al., 2014)). Between June and December 2017, 160 women were recruited from the antenatal care clinic of TTM hospital to participate in the study. To be eligible, women had four Samoan grandparents (an indicator of Samoan ethnicity), were older than 18 years, 35–41 weeks gestation, had singleton pregnancies with no complications, and resided within 30 minutes of the AUA (restricting the sample to women from AUA and NWU). Participants completed a baseline assessment at the time of recruitment and were invited to participate in 3 further follow up assessments: in the week immediately after their infant’s birth (the early infant assessment); when the infant was 2 months of age; and when the infant was 4 months of age.

For the purpose of this analysis we focused on mother-infant pairs who had completed the baseline (35–41 weeks gestation) and early infant assessment (n=114). Then, because of the rapid growth and development that occurs during the neonatal period, we further restricted the sample to n=107 pairs where the infant was less than 14 days old at the early infant assessment (mean ± SD = 4.8 ± 2.8 days; median = 4 days). A flowchart of available data and reasons for loss to follow up is shown in Figure 1. All participants gave their written informed consent before inclusion in the study and protocols were granted ethical approval by both the Yale University Institutional Review Board (HIC #2000021076) and the Samoa Ministry of Health’s Health Research committee.

Figure 1.

Flowchart of mother-infant pairs with data used for this analysis.

Questionnaires

During the baseline assessment Samoan-speaking research assistants administered a demographic survey from which we attained sociodemographic characteristics of participants (Table 1) including the Material Life Style Score, a household assets inventory, which has been used previously in Samoa to reflect socio-economic status (Chin-Hong & McGarvey, 1996).

Table 1.

Characteristics of mother-infant pairs (n=107)

Maternal Characteristics	Mean ± SD or n (%)
Census region of residence
Apia Urban Area, n (%)	54 (50.5)
Northwest Upolu (peri-urban), n (%)	53 (49.5)
Age, years	28.1 ± 5.7
1,2Material Lifestyle Score (1–15), n (%)	8.8 ± 3.8
BMI post-pregnancy, kg/m2	33.8 ± 6.7
BMI classification (Pacific Islander cut-offs)
Underweight (<18 kg/m2), n (%)	0 (0)
Normal weight (18–25.9 kg/m2), n (%)	11 (10.3)
Overweight (26–32 kg/m2), n (%)	42 (39.3)
Obese (>32 kg/m2), n (%)	54 (50.5)
Gestational age at birth, week	39.9 ± 0.9
Gravidity	1.9 ± 1.7
Current smoker at time of recruitment, n (%)	10 (9.3)
Number of cigarettes per week (smokers only)	0.3 ± 1.2 (10)
Mother’s Dietary intake per day at 31–37 weeks gestation	Mean ± SD (number of participants)
Calories	2084.9 ± 782.8
Carbohydrate, g	260.0 ± 102.6
Protein, g	81.1 ± 30.3
Fat, g	85.6 ± 37.5
Saturated Fat, g	33.3 ± 15.5
Monounsaturated Fat, g	27.7 ± 14.9
Polyunsaturated Fat, g	13.5 ± 7.0
Sugar, g	111.7 ± 52.6
Fiber, g	27.6 ± 11.6
Cholesterol, mg	328.8 ± 169.7
Sodium, mg	2185.2 ± 1277.7
Calcium, mg	614.5 ± 233.8
Potassium, mg	3522.7 ± 1241.0
Vitamin A, mg	2169.5 ± 1108.7
Vitamin E, mg	10.0 ± 4.1
3Vitamin C, mg	220.9 ± 117.0
Iron, mg	12.7 ± 5.0
Infant Characteristics
Male, n (%)	55 (51.4)
Exclusive breastfeeding, n (%)	97 (90.7)
4Birth weight, g	3496.4 ± 495.9
Macrosomia (>4000g), n (%)	15 (14)
Infant age, days	4.8 ± 2.8
Age range
1–3 days, n (%)	39 (36.4)
4–7 days, n (%)	52 (48.6)
8–14 days, n (%)	16 (15.0)
Measurements at the early infant visit
Weight, g	3462.6 ± 472.2
BMI, kg/m2	12.9 ± 1.3
zBMI	−0.4 ± 1.0
Subcutaneous fat (sum of skinfolds), mm2	19.2 ± 5.0
Abdominal circumference, mm	33.7 ± 2.2
Head circumference, mm	35.6 ± 1.1
DXA measures, total body less head (TBLH)
Fat mass, g	470.8 ± 129.9
% Fat	17.3 ± 3.0
Lean mass, g	2181.1 ± 320.5
Bone mass, g	44.6 ± 7.4
Fat-free mass, g	2225.7 ± 325.5

¹For an estimate of household characteristics, a 15-point material lifestyle score was calculated based on a summed household assets inventory using the following variables each worth 1 point: own house, fridge, freezer, stereo, portable speaker, television, VCR/DVD, couch, carpet/rugs, washing machine, landline telephone, mobile phone, computer/laptop, electricity, and motor vehicle.

²n=94, 13 participants had missing data

³n=106, one outlier was removed

⁴n=104, 3 participants had missing birth data in their medical records

Research assistants also administered a 115-item food frequency questionnaire (FFQ), which had previously been validated for use in adult Samoan populations (DiBello et al., 2009). Participants indicated the frequency with which they had consumed a fixed portion size of each food item during the last 30 days, choosing a frequency that ranged from “never/less than once per month” to “more than six times per day”. Due to the retrospective nature of the FFQ, this provided dietary data from 31–37 weeks of pregnancy. Daily nutrient intake and total energy intake were calculated by multiplying the nutrient content of a fixed portion for each food item by the mother’s daily consumption. Nutrient content information was obtained primarily from the United States Department of Agriculture’s (USDA) Food Composition Database (2018) with the Food and Agriculture Organization’s Pacific Islands Food Composition Tables (Dignan et al., 2004) providing supplementary information on six Pacific-specific foods unavailable in the USDA database.

Anthropometric measurements

Birth weight and gestational age were obtained from hospital medical records, although it should be noted that because ultrasounds were not available in most settings where antenatal care is received, medical practitioners in Samoa often estimated gestational age and rounded to 40 weeks in the birth record if a neonate appeared full term. During the early infant assessment trained research assistants collected weight and height/length measurements on the n=107 mother-infant dyads. Mothers wore light island clothing and infants were weighed in clean diapers, after taring the scale for the diaper weight. Maternal height was measured using a portable stadiometer (SECA 217, SECA, Hamburg, Germany) and weight with a digital scale (Tanita HD351, Tanita Corporation of America, IL, USA). Infant length and weight were measured with a length board (SECA 417, SECA, Hamburg, German) and digital scale (SECA 354, SECA, Hamburg, Germany), respectively. All height/length and weight measurements were measured to the nearest 0.1cm or 0.1kg and the average of the two measurements used for analysis. To calculate the variable zBMI, we used the World Health Organization standards to create age- and sex-standardised BMI z-scores (WHO Multicentre Growth Reference Study Group, 2006) (chosen for comparability to other studies of infant and child growth in this setting).

Assessment of body composition

Infant body composition was assessed using DXA. Scans were completed by one of three trained operators and analysed by one researcher (RLD). Daily quality control and quality assurance scans of a manufacturer-supplied phantom spine were performed. Infants were consistently swaddled in the same brand of blanket and scanned wearing only a clean diaper, in “thin” mode (Lunar iDXA, version Encore 17, GE Healthcare Medicine, WI, USA). The body composition outcomes generated from the total body scan included absolute measures of fat, lean, bone, and total mass (all less the head), in grams as well as percent fat, fat-free, and lean mass.

Statistical analysis

Continuous variables describing maternal and offspring characteristics are presented as means ± standard deviations, and categorical variables are presented as percentages. We first compared characteristics between the subsample of the cohort used in this analysis (n=107) and those who did not complete the early infant assessment (n=39). We also conducted sensitivity analyses to test whether there were differences in infant body composition between exclusively breastfed and mixed/formula-fed infants and infants born to mothers who smoked at the time of recruitment versus mothers who did not smoke.

After testing all outcome variables for normality, we performed bivariate analyses, examining associations of sociodemographic characteristics with both maternal nutritional intake and infant body composition outcomes to determine which covariates to include in multivariable models. We conducted independent t-tests for categorical variables and Pearson correlations for continuous variables. To control for infant age, we performed partial correlations for predictors that were significant in the Pearson correlations except for birth weight because all measurements were taken at the same time directly after birth. We then chose variables for inclusion in multivariable models based on the pattern of bivariate results and a priori hypothesised associations with infant body composition. We also explored sex differences to determine whether analyses should be sex-stratified or adjusted for sex by performing an age-adjusted general linear model analysis between the sexes for all body composition outcomes.

We then performed three types of multivariable linear regression to explore the association between maternal dietary intake and offspring body composition outcomes. First, separate residual models were created for each of the macro and micronutrients of interest. Here we controlled for total energy intake by regressing the individual nutrients on energy intake, calculating the residual and then adding a constant (Willett, Howe, & Kushi, 1997). Then we considered macronutrients with high caloric contributions (protein, carbohydrate, fat, fibre and sugar) using the substitution and partition models according to Mackerras (1996). The partition model estimates unit change in the respective offspring body composition parameter associated with a 100 kcal increase of the macronutrient of interest (non-isocaloric = total energy intake is not kept constant). The substitution model calculates the estimated unit change in the offspring body composition parameter associated with a 1% increase of energy of the macronutrient of interest and a respective 1% decrease of the other macronutrients (isocaloric = total energy intake is kept constant) and is interpreted similarly to the residual model. Total, polyunsaturated, monounsaturated and saturated fat were assigned 4 kcal per g; carbohydrates and sugar 9 kcal per g; and fiber 2 kcal per g. In all models, adjustments were made for infant sex, maternal post-pregnancy BMI, gravidity, maternal age and infant age at DXA scan (the model with birth weight as the outcome did not include infant age).

We also explored the interaction between maternal dietary intake and sex by adding an interaction term of sex by nutrient intake in all models. If the interaction term was not significant, it was removed from the final model. Where zBMI was the outcome of interest we did not include infant age at DXA scan or infant sex since those were adjusted for when we derived the zBMI score.

All analyses were performed using IBM SPSS Statistics, Version 21.0 (Released 2012. Armonk, NY: IBM Corp). No correction for multiple comparisons was made because of the exploratory nature of the study.

Results

Sample characteristics and dietary intake

Sample characteristics for mothers and infants are provided in Table 1. Women included in the study were on average 28 years old and a majority (57%) reported high school as their highest level of education. Mean postpartum BMI was 33.8 kg/m² with women predominantly categorised as having obesity (54% had a BMI over 32 kg/m²). Maternal dietary energy (kcal) and macro- and micronutrient intakes (grams, milligrams and percentages) are presented in Table 1.

Correlates of early infant body composition

There were no differences in sociodemographic or behavioural characteristics between the subsample used in this analysis (n=107) and those who did not complete the early infant assessment (n=39, see Figure 1 for details from the participant flow chart), with the exception of a biologically negligible difference in calcium intake, which was slightly greater (an additional 58.4 mg/day) in the sample used for this analysis (n=107). We found no differences in infant body composition between infants who were exclusively breastfed versus mixed/formula-fed and between groups of infants who were born to mothers who smoked versus not. We also found no effect of maternal socioeconomic status on infant body composition outcomes.

The final confounders chosen for inclusion in the multivariable models were maternal postpartum BMI, maternal age, gravidity, infant age, and sex. For models where the outcome was total subcutaneous fat, based on a sum of skinfolds, we adjusted for inter-assessor error. Maternal age was not associated with any of our outcome variables but was included in the analyses because it has previously been shown to be a predictor of infant body composition (Khalil et al., 2013).

Infant age was positively associated with subcutaneous fat (mm²) (r=0.3, p=0.005), lean mass (g) (r=0.3, p=0.007) and fat-free mass (g) (r=0.3, p=0.007) (data not shown). Gravidity was positively associated with infant lean mass (g) (r=0.2, p=0.023), fat mass (g) (r=0.2, p=0.026), and subcutaneous fat (mm²) (r=0.2, p=0.019). Postpartum maternal BMI was positively associated with infant fat mass (g) (r=0.3, p=0.002), % fat mass (r=0.3, p=0.001), bone mass (g) (r=0.3, p=0.001) and birth weight (g) (r=0.2, p=0.014). After adjusting for infant age, gravidity remained positively associations with infant lean mass (g) (r=0.2, p=0.016), fat mass (g) (r=0.3, p=0.009), and subcutaneous fat (mm²) (r=0.2, p=0.013). As did postpartum maternal BMI remain associated with infant fat mass (g) (r=0.3, p=0.001), % fat mass (r=0.3, p=0.001), and bone mass (g) (r=0.3, p=0.001).

General linear models adjusting for infant age revealed that, at the mean age of 4.8 days, female infants had 18.7% body fat compared to male infants who had 16.0% body fat (unstandardised beta= 2.7; 95% CI: 1.6, 3.7; p<0.001). Female infants had an average of 509.0g of fat mass compared to male infants who had 434.7g (unstandardised beta= 74.1; 95% CI: 26.1, 122.0; p=0.003). No other body composition measures were statistically different between sexes. Further, there were no significant sex by maternal dietary intake interactions.

Maternal nutrient intake and offspring body composition

Table 2 shows results for the residual, partition and substitution models describing the association of maternal dietary intake at 31–37 weeks gestation with bone mass (g); all other body composition outcomes were non-significantly related to maternal nutrient intake (data not presented). Table 3 presents full models, including covariate parameter estimates, for the three nutrients identified as either significantly (p<0.05) or borderline significantly (p<0.10) associated with infant bone mass in Table 2.

Table 2.

Residual, partition and substitution models of maternal diet from 32–37 weeks gestation on infant bone mass (age 0–14 days) adjusted for maternal postpartum BMI, maternal age, gravidity, infant age, and sex.

Body Composition Parameter	Nutrient	Residual Model				Partition Model				Substitution Model
		n	Beta	Std. Error	p	n	Beta	Std. Error	p	n	Beta	Std. Error	p
Bone Mass (g)	Carbohydrate	107	0.03	0.02	0.109	107	−0.01	0.01	0.067	107	0.16	0.10	0.112
	Fat	107	−0.08	0.06	0.200	107	0.01	0.01	0.221	107	−0.17	0.14	0.212
	Sat Fat	107	−0.24	0.11	0.053	107	−0.02	0.01	0.154	107	−0.52	0.27	0.058
	Mono Fat	107	−0.11	0.10	0.312	107	−0.01	0.01	0.431	107	−0.23	0.23	0.308
	Poly Fat	107	−0.09	0.25	0.714	107	−0.01	0.02	0.601	107	−0.25	0.56	0.656
	Protein	107	−0.08	0.05	0.068	107	−0.02	0.01	0.028	107	−0.35	0.22	0.112
	Sugar	107	0.01	0.03	0.556	107	0.01	0.01	0.326	107	0.08	0.12	0.534
	Fiber	107	0.22	0.10	0.032	107	0.10	0.04	0.034	107	2.25	0.99	0.025
	Cholesterol	107	0.01	0.01	0.490	---	---			---	---		---
	Sodium	107	0.00	0.01	0.852	---	---			---	---		---
	Calcium	107	−0.01	0.01	0.464	---	---			---	---		---
	Potassium	107	0.01	0.01	0.252	---	---			---	---		---
	Vitamin A	107	0.01	0.01	0.230	---	---			---	---		---
	Vitamin E	107	0.27	0.28	0.334	---	---			---	---		---
	1Vitamin C	106	0.01	0.01	0.784	---	---			---	---		---
	Iron	107	0.17	0.32	0.604	---	---			---	---		---

¹n=106, one participant with Vitamin C intake more than 2 standard deviations above the sample mean was removed from this analysis only

Table 3.

Significant multivariable linear regression models of saturated fat (borderline significant), protein and fibre on bone mass adjusted for infant age at bone mass measurement, maternal BMI, maternal age, gravidity and infant sex including covariate parameters and model type (residual, partition or substitution). n=107.

	Saturated Fat		Protein	Fibre
Predictor	Residual Model	Substitution Model	Partition Model	Residual Model	Partition Model	Substitution Model
Saturated fat, residual	−0.2 (0.1)
	(−0.5, 0.01)
	p=0.053
Saturated fat, % of kcalories		−0.5 (0.3)
		(−1.1, 0.1)
		p=0.058
Calories		0.01 (0.01)				0.01 (0.01)
		(−0.01, 0.01)				(−0.01, 0.01)
		p=0.257				p=0.251
Protein, kcal			−0.02, 0.01)
			(−0.04, −0.01)
			p=0.028
Total calories - calories from protein			0.01 (0.01)
(partition model)			(0.001, 0.01)
			p=0.022
Fiber, residual				0.2 (0.1)
				(0.1, 0.4))
				p=0.034
Total Calories - calories from fiber					−0.01 (0.01)
					(−0.01, 0.01))
					p=0.0267
Fiber, % of kcalories						2.3 (1.0)
						(0.3, 4.2)
						p=0.025
Infant age	0.5 (0.2) (0.1, 1.1) p=0.021	0.6 (0.3) (0.1, 1.1) p=0.022	0.5 (0.3) (−0.1, 1.0) p=0.063	0.6 (0.3) (0.1, 1.0) p=0.024	0.6 (0.3) (0.1, 1.1) p=0.034	0.6 (0.3) (0.1, 1.0) p=0.025
Gravidity	0.3 (0.5) (−0.6, 1.3) p=0.497	0.3 (0.5) (−0.6, 1.3) p=0.473	0.5 (0.5) (−0.4, 1.4) p=0.300	0.4 (0.5) (−0.5, 1.4) p=0.352	0.5 (0.5) (−0.5, 1.4) p=0.308	0.5 (0.5) (−0.5, 1.4) p=0.321
Maternal BMI	0.4 (0.1) (0.1, 1.1)) p<.001	0.4 (0.1) (0.2, 0.6) p<0.001	0.4 (0.1) (0.2, 0.6) p<0.001	0.4 (0.1) (0.2, 0.6) p<0.001	0.4 (0.1) (0.2, 0.6) p<0.001	0.4 (0.1) (0.2, 0.6) p<0.001
Sex (Ref: Male)	−0.8 (1.3) (−3.5, 1.8) p=0.542	−0.7 (1.4) (−3.4, 2.0) p=0.600	−0.2 (1.4) (−2.9, 2.6) p=0.905	−0.4 (1.4) (−3.1, 2.3) p=0.778	−0.4 (1.4) (−3.1, 2.3) p=0.784	−0.2 (1.4) (−2.9, 2.6) p=0.913
Maternal age (years)	−0.1 (0.2) (−0.4, 0.2)) p=0.610	−0.1 (0.1) (−0.4, 0.2) p=0.563	0.02 (0.2) (−0.3, 0.3) p=0.909	−0.1 (0.2) (−0.4, 0.2) p=0.496	−0.1 (0.2) (−0.4, 0.2) p=0.391	−0.1 (0.1) (−0.4, 0.2) p=0.411
Constant	37.6 (5.8) (26.1, 49.1)) p<.001	35.9 (5.7) (24.6, 47.2) p<0.001	27.9 (4.9) (18.3, 37.6) p<0.001	25.2 (5.4) (14.7, 35.7) p<0.001	29.5 (4.8) (20.0, 39.1) p<0.001	23.3 (5.6) (12.2, 34.4)) p<.001
Observations	107	107	107	107	107	107
R2	0.178	0.181	0.194	0.136	0.188	0.193
AIC	419.996	421.621	419.918	419.088	420.696	420.043

95% CI in parentheses; Standard error; t-statistic in brackets.

In residual models controlling for total energy intake, an increase in one gram of saturated fat and increase in one gram of fibre were associated with a 0.2g decrease (borderline significant, std. error: 0.1, p=0.053) and 0.2g increase (std. error: 0.1, p=0.032) in infant bone mass, respectively. In the partition models, an increase of 100 kcals of protein was related to decreased infant bone mass by 0.02g (std. error 0.01; p = 0.028), while an increase of 100 kcals of fibre was related to increased infant bone mass by 0.1g (std. error 0.01; p = 0.034). Finally, the substitution models, which estimate the unit change in the respective body composition parameter associated with an increase of 1% of energy of the macronutrient of interest, identified a borderline negative association with saturated fat (unstandardised beta: −0.5; std. error 0.3; p = 0.058) but a positive association with fibre (unstandardised beta: 2.3; std. error 1.0; p = 0.025) (similar to the results observed in the residual models). Regardless of the model used to examine nutritional intake, infant age and maternal BMI remained significantly and positively associated with infant bone mass, while gravidity and maternal age were no longer significant.

Discussion

To our knowledge, this is the first study to explore the association between maternal nutritional intake in late gestation and early infant body composition assessed by DXA in a population at high-risk of obesity-related non-communicable disease. Our results indicate that maternal dietary intake at 31–37 weeks gestation had an effect only on early infant bone mass, but no other body composition measures. Greater maternal intake of saturated fat and protein was inversely associated (saturated fat only borderline significantly) with bone mass, whereas fibre intake was positively associated. Although the overall influence of maternal pregnancy diet on bone mass was small, with the largest effect being a 0.2g (0.4%) increase in bone mass associated with fibre intake, this may lead to consequences later in life that we could not observe here due to the cross-sectional design of the study.

The majority of existing studies of infant body composition have focused exclusively on fat mass, likely because of the known associations of early infant fat mass and disease or the ease of collecting skinfold measurements compared to using imaging. By ignoring fat-free mass, however, researchers may be missing a key biological pathway for later metabolic health, since bone and lean mass accretion during development have been observed to be protective against metabolic disorder in adulthood (Bassett, 1994; Hong et al., 2017; Karsenty & Ferron, 2012; Odiere et al., 2010). Crume et al., (2016) used air displacement plethysmography (PEA POD) to assess neonatal body composition among 1040 infants from Aurora, USA, 72 hours after birth and showed no association between dietary intake during gestation (27 weeks) and neonatal fat-free mass. Since air displacement does not allow the partitioning of lean and bone mass, and bone mass accounts for a much smaller proportion of fat-free mass than lean tissue, any impact of maternal diet on bone mass may have been overshadowed by a lack of association with lean mass, as we observed here. The authors did, however, document a positive association between maternal saturated fat intake during the 8^th-32^nd week of pregnancy and infant fat mass (g) in a population of mostly primiparous (65%) women aged 28 years on average, with majority normal BMIs (normal, 53%; overweight and obese, 44%). We did not see this in our results, although contradictory findings are characteristic of studies of maternal nutrition and infant body composition, especially in less well-studied nutrients (opposed to folic acid which is known to play a role in fetal development) (Ramakrishnan et al., 2012).

Our findings did, however, replicate those of two other studies: one from Copenhagen, Denmark, conducted in mostly nulliparous (n=57%), obese women with a mean age of 31 years, found no association between maternal fat intake at 11–14 or 36–37 weeks and infant birth body composition measured by DXA (Renault et al., 2015), and another from Singapore with women aged 30 on average with BMIs mostly in the healthy range (normal, n=57%; overweight and obese, n=30%) found no association between maternal fat intake from 26–28 weeks gestation and birth weight (Chong et al., 2015). A key limitation of the existing literature on this topic as a whole is the comparison of populations that differ in a number of socioeconomic and human biological characteristics, including BMI. We note that Samoan women have BMIs that are 7.8 kg/m² higher than the global average, indicating abundant energy stores to support fetal growth, which may mask any immediate effects of pregnancy dietary intake on infant body composition in this and other populations with a high proportion of overweight/obesity. The longer term effects of maternal pregnancy nutrient intake though, are unknown (Christian & Stewart, 2010). Future studies that incorporate longer term data on infant body composition during the infancy or childhood period would illuminate whether maternal nutrition during pregnancy has any effect after the period of our study.

Each of the associations between dietary intake and bone mass observed here are biologically plausible, although more comprehensive and detailed mechanistic research is needed. Although our saturated fat finding was only borderline significant, higher fat diets may reflect lower dietary quality (Sharma, Greenwood, Simpson, & Cade, 2018) or lower micronutrient consumption. In animal studies, high-fat diets adversely affect bone by decreasing intestinal calcium absorption (Wohl, Loehrke, Watkins, & Zernicke, 1998; Zernicke, Salem, Barnard, & Schramm, 1995). The negative association between protein intake and bone mass could potentially be explained by the fact that in animals and humans (Cooper et al., 1996), protein intake promotes excess urinary excretion of calcium, which may limit the amount of bioavailable calcium to the fetus and affect bone growth. Finally, there is limited explanation for why fibre could affect bone mass at birth. Fibre intake, similar to saturated fat, may also represent dietary quality, however, it is also possible that fibre has indirect effects through other hormonal mediators of bone-diet interactions like the adipokine leptin which is demonstrably affected by fibre intake (Heppe et al., 2013) or through osteocalcin, which has a tentative association with fibre intake (Michaelsson et al., 1995), but about which less is known. Both hormones play a direct role in metabolism regulation and likely diabetes risk in adulthood (Considine et al., 1996; Karsenty & Oury, 2014; Murray et al., 2015) making them important targets for future mechanistic study.

Although not the primary focus of this investigation, it is important to point out both the sex differences in body composition at birth observed in our sample and the relatively higher fat mass and lower lean mass recorded among this sample of Samoan infants compared to those from other ethnic groups. The finding that girls had higher fat mass at birth replicates observations from other studies. Fields et al. (2009), for example, used PEA POD to assess body composition at 20 days postpartum among a multi-ethnic sample and observed that girls had approximately 2.4% greater body fat than boys (Girls: 15.1% ± 5.9%; Boys: 12.7% ± 4.5%). In another multi-ethnic study Butte et al. (Butte et al., 2000) used DXA and observed 2.8% greater fat mass in girls at 0.5 months of age (Girls: 14.2% ± 9.0%; Boys: 11.4% ± 8.0%). Both findings are strikingly similar in magnitude to the 2.5% difference we observed in our sample indicating that there may be inherent sex differences in tissue development during gestation. These studies both followed infants longitudinally and conflict in their conclusions about whether sex differences remain by mid-late infancy, but since early development of fat tissue may be associated with later chronic disease risk, these sex differences will be important to monitor over time among the at-risk Samoan sample. Perhaps as a result of the high level of overweight and obesity observed among the mothers included in our cohort, percent body fat in our Samoan infants was higher than observed in many other studies of early infant body composition, further warranting longitudinal analyses to determine how chronic disease risk may be conferred by early life development.

Strengths of this study include the use of DXA to measure body composition, which allows for more detailed body composition analysis than is allowed by skinfolds or air displacement plethysmography. DXA has been shown to be accurate and without bias compared to magnetic resonance imaging (MRI) in infants, which is considered one of the most accurate techniques for measuring body composition in children and adults (Fields et al., 2015). We were careful to control the DXA environment and swaddled all infants with the same blanket, limited clothing to diapers alone, and chose scans with very minimal participant movement.

Limitations of this study include the postpartum instead of pre-pregnancy maternal BMI data, relatively small sample size, and use of the FFQ to measure dietary intake. We were not able to obtain pre-pregnancy weight, which is standard for controlling for maternal BMI; many Samoan families do not own scales and prenatal care is initiated later in pregnancy than in many other settings. For these reasons we were limited to measuring maternal BMI at the time of the infant DXA. FFQ data relies on recall of dietary intake over the past month and is also limited to a pre-selected list of foods. Tools to measure dietary intake remain generally limited, but we considered the FFQ the best option for this setting. Our survey had been validated in a large sample of adult Samoans, which is different to the current study’s smaller sample of pregnant women. However, we are confident that diets of non-pregnant and pregnant adults do not vary substantially (in pregnant populations there may be avoidance of certain foods or increased consumption of fruit and vegetables (Kocher et al., 2018)), and there are no specific “pregnancy foods” that our survey would have missed. Since, the surveys were conducted in hospital and medical settings, it is possible that participants underreported eating “unhealthy” foods, a practice which has been found to be more pronounced in individuals with obesity (Maurer et al., 2006), although less stigma around obesity has been observed in the Samoan population (Brewis, 2015). Another limitation is that physical activity and gestational weight gain were not controlled for as confounders in the analysis.

While it is important to acknowledge the limitations of our study design, this work has provided precise infant body composition measurements on an ethnic minority currently experiencing extremely high rates of obesity. The information from this analysis is the first step in assessing whether nutritional interventions during pregnancy would be an effective way to reduce prevalence of early risk factors for metabolic disease in neonates (e.g., high maternal perinatal BMI, high infant birth weight or high fat mass at birth (Cnattingius et al., 2012; Leddy, Power, & Schulkin, 2008; Oken & Gillman, 2003; Yu et al., 2011)). Our results indicate that maternal diet during late pregnancy has little effect on infant adiposity from 0–14 days; but does highlight the potential impact on bone, which is often overlooked for its role in metabolic health.