Optimizing research approaches: Methodological advances in studying human milk and infant growth and body composition

Background:
Obesity and chronic disease risk has long been a public health focus in high-income settings.
This is increasingly true for low-income settings experiencing nutritional transition and an ensuing double burden of malnutrition. Nutritional exposures in early life can have lasting effects on metabolic health and chronic disease risk and alterations in growth and body composition (BC) during critical windows of development may explain some of these associations. As the earliest nutritional exposure for many infants, human milk (HM) may contribute to BC development and thus chronic disease risk. However, the study of both paediatric BC and HM are complicated by methodological difficulties, creating a challenging research environment. Bioelectrical impedance analysis (BIA) represents a feasible tool for BC assessment in low-income settings but requires population-specific calibration. HM composition is influenced by sampling and analytical protocols, and a lack of standardized tools has hindered cross-study comparisons.
Objectives:
The overall objective of this thesis was to develop improved tools for the study of HM and BC.
In paper I, the objective was to calibrate BIA for use in Ugadan children with stunted linear growth using isotope dilution as a reference method. In paper II, we aimed to apply this calibrated BIA in a large cohort of Ugandan children with stunting who were enrolled to a nutritional intervention study and investigate correlates of BC at baseline. In paper III, the objective was to produce globally representative HM macronutrient reference values (RV) for normal mothers whose infants were growing without constraint.
Methods:
This thesis is a culmination of two separate studies. The first is a nutrition intervention trial based in Eastern Uganda, to which 750 children aged 12-59 months with stunting were enrolled. We selected a purposive subsample of 50 stunted children to represent the full range of age, sex, and nutritional statuses of the main study population and developed an equation to estimate fatfree mass (FFM) from whole-body impedance. Bioimpedance was measured using a single-frequency (50 kHz) device, and deuterium dilution analysed by isotope-ratio mass spectrometry as the reference technique (paper I). In paper II the calibration equation was applied to bioimpedance measurements from the main study and multiple linear regression was used crosssectionally to compare to UK BC references and investigate baseline correlates of FFM, fat mass (FM), FFM index (FFMI), FM index (FMI), and height. Correlates were chosen based on biological plausibility and included serum inflammation markers, breastfeeding practices by recall, nutritional status, and infection markers. The second study was a multi-centre cohort including mother-infant dyads from Bangladesh, Brazil, Denmark, and The Gambia (n = 250 per site). Dyads remained in the study if they conformed to a range of health, dietary, and growth criteria so the cohort represented a normative sample. Mothers provided full-breast HM samples at three timepoints between 1 and 8.5 months, and macronutrient concentrations analysed using infrared spectroscopy. In paper III, we used generalized additive models for location, scale, and shape (GAMLSS) to produce RVs for protein, carbohydrate, fat, and energy.

Results:
In paper I we developed a BIA equation that performed with a prediction error of 4.5%. for FFM. Including HAZ in the equation explained a further 1% of variation compared to the next best model which included impedance index, sex and age. In paper II, 750 stunted children (45% female) we found deficits in absolute FM, FFM, FMI, but not FFMI, compared to UKreferences. Elevated α1-acid glycoprotein (AGP) and c-reactive protein were associated with β [95% CI] 0.34 [0.26, 0.41] kg and 0.11 [0.03, 0.19] kg lower FFM, respectively, but neither were associated with FM. AGP was further associated with 1.14 [0.76, 1.52] cm lower height and 0.15 [0.07, 0.23] kg/m2 lower FFMI. HAZ was associated inversely with FFMI, but positively with FMI. Each additional month of breastfeeding was associated with 0.02 [0.01, 0.03] kg greater FFM and 0.01 [0.00, 0.02] kg greater FM, in proportion with 0.10 [0.06, 0.13] cm greater height.A positive rapid malaria test was associated with 0.27 [0.09, 0.44] kg/m2 greater FMI, but not with FFMI (0.01 [-0.07, 0.10]). In paper III we present HM macronutrient RVs produced using 2194 samples valid for inclusion from 903 mothers at 1-8.5 months postpartum. HM protein RV median reduced from 1.15 g/100 mL at 1 month, to 0.8 g/100 mL at 4 months postpartum from where it remained stable. The RV median for HM carbohydrate, fat, and energy were also relatively stable across lactation at around 6.8 g/100 mL and 3.2 g/100 mL, and 600 kcal/L, respectively. The GAMLSS framework also accounted for changing variance, skew, and kurtosis over time.

Discussion and Conclusion:
The BIA equation produced in paper I suggests that degree of stunting changes the relationship between whole-body impedance and FFM. We propose that characteristic alterations of limb proportions in stunting could be the mediating factor, and that although the additional variance explained by HAZ was modest, it may continue along the non-stunted range of HAZ. In the main sample, we found that systemic inflammation was associated with reduced FFM and further FFMI and height for AGP. Inflammation is known to disrupt hormonal drivers of paediatric growth and may be an ideal target for interventions seeking to improve metabolic capacity associated with lean mass, while reducing metabolic load. Longer continued breastfeeding seemed protective of growth even in this wholly stunted sample of children, reinforcing the World Health Organisation recommendations to support continued breastfeeding to two years.We also believe that there is suggestive evidence of FFM protection at the expense of linear growth and FM storage in the face of stunting, indicated by a comparable FFMI compared to UK BC, and the opposing associations of FFMI and FMI with HAZ. Brain-sparring is a well-known mechanism which may extend to other lean components of BC. Our HM macronutrient RVswere produced using HM samples collected using standardised protocols from geographically and genetically diverse populations, making them globally representative and robust. We encourage their use in future research and practice as normative values, but researchers must consider their own sampling and analytical techniques used to derive HM composition estimates.