We conducted a longitudinal cohort study of 99 mother/infant dyads at the Projahnmo research site in rural Sylhet, Bangladesh. ¹⁵ The Projahnmo study site is in the northeastern area of the Sylhet district in Bangladesh. This area is rural and agrarian, with a per-person gross domestic product (GDP) of less than 2 USD daily. Breastfeeding is common, with 88% of mothers continuing to breastfeed their children until 12–23 months of age. Thirty-four percent of children under age 5 years are stunted. ¹⁶

The lactation cohort was nested within the larger Projahnmo pregnancy-birth cohort, a prospective observational study that enrolled pregnant women before 20 weeks of gestation, with pregnancy dating confirmed by enrollment ultrasound prior to 20 weeks of gestation. ¹⁵ Women were consecutively enrolled into the lactation cohort from 03 January 2018 to 04 April 2018 at the postpartum visit around 42–60 days after delivery. They were followed, with their infants, longitudinally from birth to 24 months. The procedures for this study were approved by the institutional review boards at Brigham and Women’s Hospital (#2017P000876/PHS) and the National Research Ethics Committee of Bangladesh Medical Research Council (BMRC/NREC/2016-2019/340). Written informed consent was obtained before enrollment in the lactation cohort. For lactation cohort eligibility, we included participants enrolled in the pregnancy-birth cohort who had delivered a liveborn infant and planned to exclusively or predominantly breastfeed. We excluded participants with painful breast conditions, such as mastitis or cracked nipples, or who were breastfeeding multiples.

We collected breast milk at the enrollment postpartum visit (referred to as the 2-month visit) and a follow-up visit at approximately five months. The milk was collected as a clean, hand-expressed, mid-feed sample. Participants were asked to breastfeed their infants for 3 to 5 minutes, then briefly detach the infant and hand express 15 milliliters (ml) of milk from the breast into a sterile container. The research team froze the human milk samples in 7.5 ml aliquots. The samples were transferred via cold chain to Brigham and Women’s Hospital in Boston, MA, where they were maintained at -80°C until the time of analysis.

We used mid-infrared spectroscopy to measure breast milk macronutrient content (Miris human milk analyzer, Uppsala, Sweden). This method has been validated in previously published milk composition studies and was approved by the United States Food and Drug Administration in 2018. ^{17 , 18} We used calibration and quality control procedures per manufacturer recommendations to ensure that measured values remained within an acceptable accuracy range. ¹⁸ The Miris reports crude protein, fat, and carbohydrate values. Crude protein is determined by estimating the total nitrogen content of a sample, which is then converted by a factor of 6.38. True protein is then determined as 0.8*crude protein. ^{17 , 18} We used true protein concentrations for our analyses. We determined metabolizable energy concentrations as a calculated value using the Atwater general equation (true protein [g/100ml]*4 + carbohydrate [g/100ml]*4 + fat [g/100ml]*9 = energy). ¹⁹ Metabolizable energy is more clinically relevant than the gross energy values reported by the Miris analyzer. ²⁰

At both the 2- and 5-month lactation visits, we collected breastfeeding diaries in which participating mothers documented the timing and duration of every breast feed in 24 hours. Study nurses also collected a single pre- and post-feed test weight at the study visit measured on a TANITA BD585 Pediatric Scale (precision: 10 grams [g]), which was used as a proxy for breast milk intake volume. ²¹ We multiplied the volume of the observed single feed (pre-post weight) by the number of feeds in the prior 24 hours collected by lactation diary to estimate the total daily breast milk intake. We then multiplied the estimated volume of breast milk per day by the breast milk macronutrient and energy concentrations to approximate the daily intake of macronutrients and energy. ²¹

Infant anthropometrics, including weight, length, and head circumference, were obtained using Intergrowth-21st standardized procedures at birth, 3, 6, and 12 months. ²² We measured infant weight using digital scales (TANITA BD585 Pediatric Scale; precision: 10 grams [g]), length using a length board (Seca 417 portable infantometer and Seca 213 mobile stadiometer; precision: 1 millimeter [mm]), and head circumference using insertion tape (JiVitA measuring tape; precision: 1 mm). All measurements were obtained in triplicate, and the median value was used.

Infant birth z-scores were calculated for gestational age and sex using the Intergrowth-21st growth standards for comparison of birth measures among infants on a common scale. ^{22 , 23} Small-for-gestational age (SGA) was defined as a birthweight below the 10th percentile for gestational age and sex. Infant 3-, 6-, and 12-month z-scores were calculated using the WHO sex-specific growth curves. ²⁴ Our primary outcome was length-for-age z-score at 6 months. Secondary outcomes included length-for-age z-score at 3 and 12 months, head circumference-for-age z-score at 3, 6, and 12 months, and weight-for-age z-score at 3, 6, and 12 months.

We assessed child neurodevelopment at 24 months using the Bayley-III. ²⁵ The Bayley-III measures language (receptive and expressive), motor (gross and fine), and cognition. We used a version of the Bayley-III that has been adapted and validated in rural Bangladesh. ^{26 – 29} Cultural adaptation of the Bayley-III primarily consisted of picture modification to be appropriate to Bangladesh while maintaining the original question intent. The Bayley-III was administered by study staff with a master’s degree who had received a three-week, rigorous training by a senior psychologist. Data collection was initiated only after attaining at least 95% agreement between testers and trainer (r = 0.95, p ≤ 0.001) for language, motor, and cognitive domains. ³⁰ Based on our team’s prior experience and studies in Bangladesh, we chose to standardize raw Bayley-III scales by generating z-scores within the larger Projahnmo cohort for these analyses instead of using age-standardized conversion tables based on Western population norms. ^{31 , 32} Our primary neurodevelopment outcome was the Bayley-III z-score in the cognitive domain. Secondary outcomes included language and motor z-scores.

We summarized descriptive statistics for continuous variables as either means and standard deviations (SD) for normally distributed data or median and interquartile ranges (IQR) for non-normally distributed data. Categorical variables were reported as frequencies and percentages. We categorized macronutrient concentrations (g/100ml or kcal/oz) and daily intake (g/day or kcal/day) into low, medium, and high tertiles at each timepoint. Exposures were assessed as low versus high (reference) and middle versus high (reference) tertiles. Differences in continuous concentration and intake measures at 2- and 5-months were tested using Wilcoxon rank sum test. We conducted univariate and multivariable linear mixed regression to quantify the mean difference in infant weight-for-age z-score, length-for-age z-score, and head circumference-for-age z-scores when comparing the lowest tertile of breast milk concentration and volumes to the highest tertile reference group. Mixed models accounted for repeated measures within the same subject and included fixed effects for exposures (categorical macronutrient or energy intake or concentration), time of growth measurement (age in months), and the interaction between the exposure and time. Based on our conceptual model in Figure 1, multivariable models were adjusted for potential confounders determined a priori including wealth index, maternal age, maternal BMI, and birth z-score. To account for individual differences, we specified a random intercept for each infant with an identity covariance structure. To model the correlation between repeated measurements within the same infant, we specified an exchangeable residual covariance structure, which assumes equal variances at both time points and constant correlation between measurements within the same child. The models were fitted using restricted maximum likelihood estimation with repeated degrees of freedom. We report calculated associations as mean differences (MD) or adjusted mean differences (aMD) and 95% confidence intervals (CI).

For neurodevelopment, we used univariate and multivariate linear regression to determine the association between macronutrient and energy concentration and intake tertiles (exposure) and Bayley-III z-score for language, cognitive, and motor domains (outcome) at 24 months, comparing infants in the lowest tertile to the highest tertile reference group. Based upon our conceptual model ( Figure 1), multivariable models were adjusted for potential confounders determined a priori: wealth index, maternal education, maternal age, and weeks preterm. Analyses were not adjusted for multiple comparisons. Analyses were performed using Stata software, v18 (StataCorp LLC., College Station, TX, USA). ³³

We reported these associations as beta coefficients, interpreted as the mean difference (MD) between binary exposure group status, and 95% CIs. We reported our methods and results per the Strengthening the Reporting of Observational Studies in Epidemiology-Nutritional Epidemiology (STROBE-nut) reporting guidelines. ³⁴

Infants with low protein intake at 5 months had lower length z-scores at 12 months (aMD -0.51 [-0.98, -0.05]). There was a consistent, but not significant, trend of lower weight z-scores at every time point among participants with lower carbohydrate intake at both 2- and 5-months. Adjusted associations between daily intake and growth are shown in Table 3. Unadjusted results are shown in Supplementary Table 2.

Infants exposed to low energy concentrations at 2 months had higher head circumference z-scores at 3 and 6 months compared to the high energy concentration group (3 months: aMD 0.51 [95% CI 0.10, 0.92]; 6 months: aMD 0.43 [0.06, 0.80]). Low fat concentrations at 2 months were associated with higher weight z-scores at 12 months (aMD 0.50 [0.04, 0.97] g/100ml). There were no other significant associations between macronutrient or energy concentrations and infant growth z-scores after adjusting for confounders ( Table 4). Unadjusted results are shown in Supplementary Table 3.

Infants with low protein intake at 2 months had higher 24-month Bayley-III language z-scores than those with high intake (aMD 0.51 [0.02, 1.02]). Infants with low energy intake at 5 months had adjusted 24-month cognitive scores that were 0.61 z-scores lower than those with high intake (95% CI -1.22,-0.00), with a nonsignificant trend toward lower motor scores (aMD -0.54 [-1.11, 0.02]). Infants with low carbohydrate intake at 5 months had 24-month motor scores that were 0.65 z-scores lower than those with high intake (95% CI -1.20, -0.08). Adjusted associations are shown in Table 5 and unadjusted associations are shown in Supplementary Table 4.

In our cohort, breast milk macronutrient and energy concentration tertiles at 2 and 5 months were not associated with Bayley-III cognitive, language, or motor z-scores at 24 months (Supplementary Tables 5-6).

Infants in our cohort with the lowest protein intake at 5 months had lower length z-scores at 6 months and 12 months. Five-month protein concentrations demonstrated a similar trend, which was non-significant. Cisse et al and Sims et al found similar results to ours in Senegal and the United States, respectively, with higher breast milk protein intake associated with increased length-for-age Z-scores at 3 months and up to 9 months. ^{39 , 40} Other studies have found no associations between daily protein intake and infant growth. ^{41 – 43} Notably, the studies that found no association between protein intake and length were all very small (n ≤ 30 infants), so they may have been underpowered to detect this outcome.

We found no association between breast milk protein concentrations at 2 or 5 months and infant growth. A small number of other LMIC-based studies have examined the association between breast milk protein concentration and growth. These studies found mixed results. A study in Egypt of hind milk collected at 6-14 months postpartum found no correlation between breast milk macronutrients and infant weight or length. ⁵ However, a larger study of 40 infants in Indonesia found higher protein concentration correlated with improved head circumference and linear growth, though only at the first month. ⁴⁴ Overall, studies on the relationship between breast milk protein concentrations and infant growth have been mixed.

Although fat is one of the most variable components of breast milk and is the largest contributor to overall caloric intake and density, there was no association between either fat intake or concentrations and infant growth in our cohort. This is consistent with most previous studies, which have similarly found little association between breast milk fat and infant growth, suggesting that other factors related to nutrient intake or infant physiology are more critical determinants than fat in young infant growth. ⁴

We found a consistent trend of higher carbohydrate intake at 2- and 5-months associated with higher weight Z-scores as compared the low or medium intake groups at all timepoints. Two out of four studies in the Brockway review also found a positive association between carbohydrate intake and infant weight, including a study of Australian infants which observed this association at 2, 5, 9, and 12 months of age. ^{4 , 45} While a clear link exists between dietary sugar intake and weight gain among older individuals, it seems that high dietary breast milk carbohydrate intake, in the form of lactose and the minimally-digested human milk oligosaccharides, may also be seen with higher weight gain parameters among breastfeeding infants.

Infants consuming breast milk with lower concentrations of energy at 2 months and fat at 5 months had higher 3-month head circumference Z-scores and 12-month weight Z-scores, respectively. Interestingly, these associations were not significant after accounting for volumes in our analysis of daily energy and fat intake. Our findings are consistent with studies from both England and Turkey that have demonstrated an inverse relationship between breast milk fat percentage and infant weight gain. ^{37 , 46} The Turkish study instead found a positive association between volume intake and weight gain. A longitudinal study in the Netherlands demonstrated that infants who consume breast milk with higher caloric content show earlier signs of satiety and finish feeding more quickly than infants consuming lower-calorie breast milk. ⁴⁷ Overall, our findings are consistent with the biologically plausible idea that intrinsic hunger-satiety signaling may lead infants to titrate their volume intake to account for higher or lower energy concentrations in breast milk.

We found multiple associations between total daily intake of breast milk macronutrients or energy and infant neurodevelopment. At 5 months, the infants in the lowest carbohydrate and energy intake tertiles had lower motor and cognition scores, respectively, than those in the highest tertile. Surprisingly, infants in the lowest 2-month protein intake group performed better on the language domain than those in the highest group, though this was not a primary hypothesis. It is difficult to compare these findings to prior literature, as nearly all previous studies have evaluated infants born preterm in high-income countries. By contrast, our study was comprised of mostly full-term infants. These studies have demonstrated that higher daily intake of protein, fat, and energy is associated with improved brain growth at term-corrected age among very preterm infants as measured by magnetic resonance imaging (MRI). ^{48 , 49} However, the effects of higher macronutrient and energy intake on neurobehavior have been less conclusive, as evidenced by a recent systematic review, which concluded there is insufficient evidence to state that there is a clear association between early energy intake for preterm infants and scores on the Bayley scale. ⁵⁰ Overall, previous studies on the relationship between breast milk macronutrient and energy intake and term infant neurodevelopment have been limited in number and scope.

Our study has certain limitations. Given the cultural context, we were only able to collect mid-feed breast milk specimens. Though a full expression would have been optimal, at the time of the study it was not culturally acceptable. However, our use of a mid-feed sample to measure macronutrient concentrations is consistent with other studies and was demonstrated in a systematic review to approximate the values of a full-feed sample. ⁵¹ A further limitation was the observational nature of our study. There are many postnatal factors that may affect growth and development during infancy and early childhood. Though we accounted for some of these in our models, there may be unaccounted-for factors that could have affected child outcomes.

Our study had many strengths. Most previous studies of macronutrient concentrations and growth or neurodevelopment in LMICs have limited their evaluations to a single time point or a single macronutrient. ⁴ Our data are enriched by studying all macronutrients and energy within a single cohort at two time points in lactation, allowing us to explore the biology of breast milk over time. We found that daily macronutrient intake to be associated with many more metrics of growth and neurodevelopment than concentrations alone. Our study is also strengthened by accurate gestational age dating by ultrasound between 8 and 19 weeks of gestation. This is unusual in rural Bangladesh, where access to ultrasound-based gestational age determination is limited. Accurate gestational age is important in order to assign growth z-scores based on postmenstrual age. Finally, the Bayley-III was administered by highly skilled study psychologists who had undergone rigorous training using a culturally-relevant, validated version of the Bayley-III.