The association between plasma choline, growth and neurodevelopment among Malawian children aged 6–15 months enroled in an egg intervention trial

Choline is an essential micronutrient that may influence growth and development; however, few studies have examined postnatal choline status and children’s growth and development in low- and middle-income countries. The aim of this observational analysis was to examine associations of plasma choline with growth and development among Malawian children aged 6–15 months enrolled in an egg intervention trial. Plasma choline and related metabolites (betaine, dimethylglycine and trimethylamine N-oxide) were measured at baseline and 6-month follow-up, along with anthropometric (length, weight, head circumference) and developmental assessments (the Malawi Developmental Assessment Tool [MDAT], the Infant Orienting with Attention task [IOWA], a visual paired comparison [VPC] task and an elicited imitation [EI] task). In cross-sectional covariate-adjusted models, each 1 SD higher plasma choline was associated with lower length-for-age z-score (−0.09 SD [95% confidence interval, CI −0.17 to −0.01]), slower IOWA response time (8.84 ms [1.66–16.03]) and faster processing speed on the VPC task (−203.5 ms [−366.2 to −40.7]). In predictive models, baseline plasma choline was negatively associated with MDAT fine motor z-score at 6-month follow-up (−0.13 SD [−0.22 to −0.04]). There were no other significant associations of plasma choline with child measures. Similarly, associations of choline metabolites with growth and development were null except higher trimethylamine N-oxide was associated with slower information processing on the VPC task and higher memory scores on the EI task. In this cohort of children with low dietary choline intake, we conclude that there were no strong or consistent associations between plasma choline and growth and development. The Mazira Project randomised trial (clinicaltrials.gov: {“type”:”clinical-trial”,”attrs”:{“text”:”NCT03385252″,”term_id”:”NCT03385252″}}NCT03385252) took place in rural Malawi from February 2018 to January 2019. This trial investigated the effect of providing one egg per day versus a nonintervention control among 660 Malawian children. Children aged 6–9 months were individually randomised to intervention or control for 6 months. The intervention group received weekly batches of eggs, and caregivers were asked to feed the child one egg per day in addition to normal feeding. The control group received no eggs, and caregivers were asked to feed the child as they normally would. Both groups received twice‐weekly home visits, as well as information about food hygiene and handwashing. Descriptions of baseline characteristics by group are reported elsewhere (Stewart et al., 2019). Briefly, baseline egg consumption was low and similar between groups (4.0% in the control group and 4.2% in the egg group). Children residing in the catchment areas of two health centres (Lungwena Health Center and St. Martins Rural Hospital in Malindi) were eligible to enrol. These areas are rural, with most families engaged in fishing and agricultural labour. Staff recruited age‐eligible children and caregivers during household visits. Exclusion criteria were: egg allergy, history of serious allergic reactions, congenital defects or conditions which may affect growth and development, severe anaemia (haemoglobin <5 g/dl), low mid‐upper arm circumference (<12.5 cm), presence of bipedal oedema or acute illness or injury warranting hospital referral. Children of families who planned to leave the study area within the next 6 months were also excluded. Detailed descriptions of data collection for this trial have been previously published (Prado et al., 2020; Stewart et al., 2019). Briefly, children and caregivers came to the study site at enrolment and at study end 6 months later. At both times, staff collected anthropometric, dietary, demographic and developmental data. Blood samples were collected to assess exclusion criteria, including tests for haemoglobin concentration (Hemocue 201, HemoCue Inc.) and malaria antigen (DF Bioline Malaria Ag P.f/Pan, Abbott Diagnostics). At the 6‐month follow‐up, the Family Care Indicators (FCI) interview was administered, which assesses children's opportunities for stimulation (Hamadani et al., 2010). Other data were collected during home visits. Soon after enrolment, staff administered the Household Food Insecurity Access Scale (Coates et al., 2007) and Home Observation Measurement of the Environment (HOME) Inventory (Caldwell & Bradley, 2003) questionnaires and collected data on housing materials and animal ownership for incorporation into a housing and asset index. After 3 months of enrolment, study staff collected anthropometric and dietary data during a home visit. Throughout the study, caregivers reported weekly on child morbidity symptoms, including the number of days with diarrhoea. The longitudinal prevalence of diarrhoea was calculated as the number of days with reported diarrhoea divided by the total number of days of recall. Trained and standardised pairs of anthropometrists measured children's recumbent length (in cm) using a Holtain length board, weight (in kg) using a Seca 874 digital scale, and head circumference (in cm) using insertion tapes (Health Books International at enrolment and Seca model 212 at 6‐month follow‐up). World Health Organization Growth Standards were used to convert values to z‐scores (length‐for‐age [LAZ], weight‐for‐age [WAZ], weight‐for‐length [WLZ] and head circumference‐for‐age [HCAZ]) (WHO Multicentre Growth Reference Study Group, 2006). In addition to continuous z‐scores, conditional and dichotomous measures were calculated, in line with best practices for linear growth analyses (Wit et al., 2017). To calculate conditional variables, anthropometric data collected at 6‐month follow‐up was regressed on data from enrolment and 3‐month follow‐up (Stein et al., 2010). The residuals reflect how each child's growth over the 6‐month study period differed from expected based on their initial growth status compared to the other study participants. A positive value reflects comparatively faster growth (or slower faltering); a negative value reflects comparatively slower growth (or quicker faltering). Because the insertion tapes were changed during the study, a conditional measure of head circumference was not included. Dichotomous outcomes were defined as being above or below a cutoff: stunted (LAZ ≤ −2), underweight (WAZ ≤ −2), wasted (WLZ ≤ −2) or small head circumference (HCAZ ≤ −2). Four developmental assessments (two behavioural measures and two measures based on eye‐tracking) were conducted by trained and standardised data collectors. The Malawi Developmental Assessment Tool (MDAT) includes 136 items across four domains (fine motor, gross motor, personal social and language development) which are scored as pass/fail based on the child's performance (or, for the personal social domain, parental report). For each domain, z‐scores were calculated based on published Malawian norms (Gladstone et al., 2010). The MDAT has been validated for use in Malawi and has high sensitivity (97%) and specificity (82%) to detect neurodevelopmental impairment in this context (Gladstone et al., 2010). In the elicited imitation task, children demonstrate declarative memory by imitating sequences of actions demonstrated by outcome assessors (Bauer, 2010). In our adaptation of the task, children were asked to imitate eight two‐action sequences (16 target actions, 8 ordered sequences) performed using sets of toys. For each set, assessors first recorded which actions the child performed before the demonstration (spontaneous actions) during a 30‐s free play. Then, the assessor performed the sequence twice while the child watched. Afterwards, the assessor scored the child's ability to reproduce the target actions (actions recalled score: 0–16) and sequences (sequences recalled score: 0–8) during two 30‐s imitation sessions. Information on the adaptation and piloting of this task has been published (Prado et al., 2020). In a few cases, children were unable to complete items due to fussiness, sleepiness, or missing or damaged toys; to correct for this, scores were calculated by multiplying the percent of correct actions or sequences available to the child by the maximum possible score. Children who were offered fewer than 8 actions or 4 sequences were excluded. The elicited imitation task was measured at 6‐month follow‐up only. Children were seated on their caregiver's lap facing a monitor with an eye‐tracking system (Tobii Pro X2‐60) attached. The eye tracker recorded the x and y coordinates of the child's gaze on the screen 60 times per second by calculating the position of the pupil and corneal reflection. Two eye‐tracking stations were used at the study site, each including a laptop, monitor, webcam and eye tracker surrounded by 4 black curtains. At baseline, children who enroled before 4 April 2018 (39%) completed a pilot version of the eye‐tracking tasks, which was later revised; these children were not included in analysis of baseline eye‐tracking data. The visual paired comparison (VPC) task measures children's recognition memory based on the concept of novelty preference, or children's preference to look at unfamiliar items. In this version of the test based on a previous study (Rose, 1983), children were presented with two identical stimuli (an African face) on the left and right sides of the screen for 20 s (the familiarisation period). Then, after a brief delay, the familiar face was shown on one side of the screen paired with a novel face on the other for 20 s, with the position of the faces reversed at 10 s (the recognition memory period). This was repeated with different faces four times. Two outcome measures were calculated from this data: a novelty preference score and the peak look length during familiarisation. To create novelty preference scores, the number and length of fixations to each side of the screen were calculated using the Tobii I‐VT fixation filter. Fixations represent a period when infants' gaze position is stable and directed towards a specific focal point. For each trial, the novelty preference score is the percent of time spent fixating on the side of the screen containing the novel stimulus compared to the total time looking at the screen. Trials with <1 s of looking time during the familiarisation or recognition memory periods were excluded (11% of trials) since the child may not have been on‐task. Peak look length was calculated as the duration of the longest look during the familiarisation phase of each trial. In previous studies, shorter looks were associated with improved attention and faster information processing (Frick et al., 1999); however, these studies used human scorers rather than eye‐tracking devices. To mimic the ability of human scorers to identify eye movements, fixations identified by the Tobii filter were recoded into ‘looks,’ which were defined as periods of visual attention towards one side of the screen that lasted ≥1 s and were not interrupted for longer than 1 s. In the Infant Orienting with Attention (IOWA) task, children demonstrate their attentional processes by shifting their gaze towards targets appearing on the screen (Ross‐Sheehy et al., 2015). In this task, children were shown a central image (a smiley face), then a 100‐ms visual cue (a small black circle), followed by a 1000‐ms target (a picture of a colourful everyday object) on one side of the screen. Children's gaze was tracked as it shifted from the central image to the target, and the response time was defined as the time from the appearance of the target to the first fixation on that side of the screen. Trials with <200 ms of looking time at the central image were excluded (10% of trials), as the child was not properly fixated on the centre of the screen. Trials with response times <100 ms or >1000 ms were also excluded (1% of trials), as they may reflect eye movements that started before the appearance of the target or off‐track behaviour, respectively. The IOWA task included 96 trials across four conditions which varied by the location of the visual cue (same side as target, opposite side, both sides or not present). Venous blood was collected into lithium heparin tubes at baseline and 6‐month follow‐up. Samples were centrifuged within a mean of 28 (SD 42) min of collection. Plasma and cell samples were separated into aliquots, which were stored in the local freezer at −20°C within a mean of 37 (SD 14) min of centrifugation. Each afternoon, the aliquots were transported to the main laboratory for storage at −80°C. Details of plasma choline measurement for this study have been described (Bragg et al., 2022). Briefly, plasma choline was measured at baseline and 6‐month follow‐up using two analysis methods. First, plasma choline was measured in a subsample of 400 children using ultra‐high performance liquid chromatography–tandem mass spectrometry (UPLC‐MS/MS) by Metabolon Inc. These semi‐quantitative data describe the distribution of plasma choline concentration and may be used for regression analysis; however, they are in relative intensity units. Betaine, DMG, TMAO and more than 800 other metabolites were also measured in this way. Additionally, plasma choline was measured quantitatively in a subsample of 60 children using liquid chromatography–tandem mass spectrometry (LC‐MS/MS) at the USDA Western Human Nutrition Research Center. These data provide the absolute concentration and can be used to compare to other studies; however, due to cost restraints, the sample size was limited, and these data were not used in regression analyses. Betaine and TMAO, but not DMG, were measured using similar validated and standardised protocols. Plasma concentrations using the two different methods were well correlated (choline: r = 0.92, betaine: r = 0.98, TMAO: r = 0.98; Bragg et al., 2022). Other metabolites measured include plasma docosahexaenoic acid (DHA), leucine, C‐reactive protein (CRP), alpha(1)‐acid glycoprotein (AGP), ferritin and zinc. Plasma DHA and leucine were measured in relative intensity units using UPLC‐MS/MS by Metabolon Inc. CRP, AGP and ferritin were measured using enzyme‐linked immunoassay by the VitMin lab (Erhardt et al., 2004). Ferritin was adjusted for inflammation, as measured by CRP and AGP, using the Biomarkers Reflecting Inflammation and Nutritional Determinants of Anaemia (BRINDA) approach (Namaste et al., 2017). Plasma zinc was measured using inductively coupled plasma mass spectrometry at Washington University in St Louis. The sample size for the randomised trial was 660 children, based on the hypothesised difference in LAZ between groups at the end of the intervention, which was the pre‐specified primary outcome of the trial. This analysis included a subset of 400 children (200 per group) who provided adequate blood samples at baseline and 6‐month follow‐up and were randomly selected for semi‐quantitative (UPLC‐MS/MS) lab analysis. This sample size is sufficient to detect correlations between plasma metabolites and growth and developmental outcomes as small as 0.14 with 80% power and a two‐sided α of 0.05. Statistical analysis plans were developed and shared publicly before analysis (https://osf.io/vfrg7/). To examine cross‐sectional associations between plasma choline and continuous growth and developmental outcomes, we pooled data from both time points and used plasma choline at the corresponding time point as a predictor in linear regression models, with time point as a covariate. Similarly, dichotomous outcomes were examined using logistic models with both time points. We used robust standard errors with participant as the independent unit to account for repeated measures within participant over time and multiple trials per participant for certain developmental outcomes (VPC novelty preference and peak look length, IOWA response time). To examine predictive associations of baseline plasma choline with growth and developmental measures at 6‐month follow‐up, we ran linear regression models with baseline plasma choline as a predictor. Each model was assessed for linearity, outliers and normality and homoscedasticity of residuals. Children with missing plasma, growth or developmental data were not included. All analyses are reported as the difference in growth or developmental outcome per 1 SD higher plasma choline, except logistic models, for which odds ratios are reported. Minimally adjusted models included variables related to data collection, specifically: time of last food intake other than breast milk, water, or tea before blood draw, calendar month of blood draw and anthropometrist or developmental data assessor. Because eggs are rich in choline, group assignment was included; however, previous analyses show that plasma choline values were similar between groups at baseline and were not statistically significantly different between groups at the 6‐month follow‐up (Bragg et al., 2022). The time point of data collection was included in models that contained both time points. For developmental assessments, the child’s mood, activity level and interaction with the assessor during tasks were included. For the elicited imitation actions recalled score, the ‘spontaneous actions’ score was included to account for actions performed independent of memory. For eye‐tracking tasks, an eye‐tracking station indicator was included. For the novelty preference score, the total time spent fixating on the screen during the familiarisation period was included. For IOWA response time, condition was included. Covariates for fully adjusted models were pre‐specified based on a theoretical causal model framework, with the following variables assessed for inclusion: child age, sex and birth order, baseline maternal age and education category, baseline household asset index and baseline food insecurity category. For growth outcomes, additional covariates included plasma leucine and zinc; plasma inflammatory markers (CRP and AGP); longitudinal prevalence of diarrhoea; and maternal height. For developmental outcomes, additional covariates included baseline plasma inflammation‐adjusted ferritin and child stimulation (as measured by HOME at baseline and/or FCI at 6‐month follow‐up). Plasma DHA was not included as a covariate, as a portion of choline’s effect on neurodevelopment may be via synergy with DHA (Mun et al., 2019). Breastfeeding may affect plasma choline levels, neurodevelopment and growth; however, it was not included in our statistical models due to nearly universal prevalence of breastfeeding (>99%) in the study sample throughout the study period. Variables associated with the growth or developmental outcome with p < 0.1 were retained in the final multivariable models. Additionally, we tested the association of choline metabolites betaine, DMG and TMAO with growth and development in minimally adjusted, exploratory models, using similar methods as described for plasma choline. Finally, we tested potential effect modifiers of the associations between choline, growth and development in exploratory minimally adjusted models. All analyses used two‐sided tests with an alpha of 0.05. Given the large number of exploratory tests, significant p values should be interpreted cautiously.