This is a randomized controlled feeding trial with four groups. After screening, eligible participants and their caregiver(s) will visit the Clinical & Translational Research Center (CTRC) at Children's Hospital Colorado (CHCO) to complete the baseline visit (Figure 1) at 5 months of age. After obtaining informed consent, participants will be assigned and randomized to one of the four study diets: (1) a meat-protein-predominant diet group (Meat); (2) a dairy-protein-predominant diet group (Dairy); (3) a plant-protein-predominant diet group (Plant); or (4) a reference group without intervention (Reference).

The baseline visit includes consenting to study participation, a questionnaire of infant feeding and family health history, family demographics, parental weight and height, gestational weight gain, parity, maternal smoking, and other variables that could affect outcomes. Blood, urine and stool samples are collected. Urine samples are used for body composition and total energy expenditure assessments using the doubly labeled water method with procedures designed for infants. A fasting breastmilk sample is also collected at baseline. When the intervention ends at 12 months of age, participants come to Children's Hospital Colorado CTRC to complete the end of the intervention visit. During the intervention (5–12 months), monthly home visits are conducted to (1) deliver study foods to Meat, Dairy and Plant and grocery vouchers to Reference, and infant formula if formula-fed or mixed feeding; (2) complete a health and dietary questionnaire; (3) obtain length and weight measurements; (4) collect a stool sample; (5) collect 3-day diet record and the food tracking log; (6) collect a breastmilk sample if breastfeeding.

Participants are being recruited in the metro Denver area from households with 3–4 month-old infants via direct mailing by Colorado Department of Public Health & Environment (CDPHE) which has access to the birth registry. Up to 300 infants will be recruited (75/group) to have at least 240 completers (60/group), based on a conservative estimate of 20% attrition.

Ethical approval for this clinical trial was obtained through the Colorado Multiple Institutional Review Board (COMIRB) (COMIRB 20-2232). COMIRB conducts the initial and annual reviews of the trial, monitoring enrollment and retention, protocol deviations, and any reported adverse effects as a result of the study. All participants are provided with a copy of the signed informed consent. This protocol is registered at ClinicalTrials.gov NCT05012930.

Participants in the three intervention groups are asked to avoid protein-rich foods from other assigned groups (30) and consume protein only from their assigned group. All complementary foods, including infant cereal, low-protein fruits and vegetable purees, and designated protein-rich foods are provided to the three intervention groups. For formula-fed infants, the same brand cow-milk based infant formula is provided. The most recent NHANES report showed that the median protein intake of US infants 6–11 months (31) is 10% total energy or 2.5 g/kg/d. This quantity is used as the targeted total protein intake. Caregivers of infants in the reference/control group are given compensation to purchase complementary foods. Based on weight recorded at the visit and the reported formula/breastmilk intake, a recommended amount of meat- dairy- or plant-based food will be provided to caregivers to approximate a total protein intake of 2.5 g/kg/d. This estimation is updated every month. A sample diet plan is in Table 1, based on a formula-fed infant with a 60th percentile weight-for-age at 9 months.

Monthly home visits at ages 6, 7, 8, 9, 10, and 11 months of age (±7 days) are conducted for each participant by one or two study coordinators. At each home visit, coordinators collect the following: stool sample (pre-collected by parents within 48 h before the visit), anthropometric measures (weight, length, head circumference), health questionnaire, 3-day diet record, food intake log, and breastmilk collection if the participant consumes breastmilk. The 9-month diet record is a weighed 3-day diet record. Caregivers also conduct test weighing again at 9 months if the participants consume breastmilk. Study coordinators also deliver study food (intervention groups) or food voucher (reference group) at the visit. Growth parameters are plotted on the infant's growth chart and Z-scores are calculated.

At 12 months of age (±7 days), the study participant returns to the Children's Hospital CTRC to complete their end of intervention visit. The visit includes the same procedures and sample collections as the baseline visit. The dietary intervention ends at 12 months. At the 12-month visit, caregivers are asked whether they want to consent to a 24-month follow-up visit. If so, caregivers sign another consent for the 24-month follow-up. When the participant turns 24 months, 1 year after the intervention ends, they come back to the Children's Hospital CTRC and repeat the procedures at baseline and 12 months, except for body composition and total energy expenditure assessment.

1) Infant growth

Pediatric nurses at the Children's Hospital Colorado CTRC (at baseline, 12 and 24 months) and study coordinators (at monthly home visits) obtain a series of infant anthropometric measurements. These measurements including weight (Seca^TM 337 infant scale), length (obtained in recumbent position using an infant stadiometer), and head circumference-for-age (Seca infant head circumference measuring tape). Growth Z scores are calculated using length and weight based on WHO/CDC standards (32).

2) Body composition and total energy expenditure at 5 and 12 months of age

At baseline and 12 months, a pre-dose urine sample is collected to document basal enrichment of ²H and ¹⁸O, then 0.1 g/kg body weight 99% enriched ²H₂O and 0.3 g/kg 10% enriched H218O are orally administered. Caregivers are asked to collect one daily sample for 7 days post-dosing. Urine samples are analyzed for ¹⁸O and ²H enrichment by Off-Axis Integrated Cavity Output Spectroscopy (OA-ICOS, Los Gatos Research Inc., Mountain View CA) (33). Isotope dilution space will be calculated via the standard equation of pre-dose and zero-time intercept enrichment via back-extrapolation, which is validated in infants (34). Total body water will be calculated from dilution space × 1.04 to account for water exchange in non-aqueous tissues. Finally, total fat-free mass will be obtained from total body water ÷ age/sex-specific hydration factor (i.e., 79% for 6-month-old infants). CO₂ production, obtained from the fractional turnover rates of ²H and ¹⁸O (35), is converted to total energy expenditure via Weir equation (36).

3) Gut microbiota

Caregivers will obtain a fresh stool sample each month of participation (5–12 months of age) and one at 24 months. 16S rRNA gene amplicon sequencing (37–39) will be performed on all samples to profile fecal bacterial communities. Following sequence QA/QC (37–39), 16S rRNA gene sequences are classified using the SINA/Silva classifier.

Shotgun metagenomic sequencing (40) will be performed on baseline and 12-month samples to determine the functional genomic capacity of fecal microbiota and broaden taxonomic profiles to include archaea, bacteriophage, DNA viruses, and microbial eukaryotes. Following sequence QA/QC, [1–4] sequences will be co-assembled with MEGAHIT [5], open reading frames predicted (Prodigal [6]), and annotated using DIAMOND [7] to query the NCBI non-redundant database. Bacterial species and strains are inferred using MIDAS [8].

All sequences and associated metadata will be deposited in the NCBI sequence read archive, following the MIMARKS standard.

1) Dietary intake

Three-day diet records are analyzed by the CTRC Nutrition Core using the NDSR software (Minneapolis, MN) by a trained dietitian. The Food and Nutrient Database in NDSR contains over 18,000 foods, including brand-name infant foods used in the proposed study. Outputs include total energy, carbohydrate (fiber, fructose, glucose, sucrose, starch, etc.), fat (cholesterol, different kinds of fatty acids), protein (meat, dairy, plant, breakdown of amino acids), vitamins and minerals. Macronutrient and energy profiles of breastmilk samples are assessed via mid-infrared spectroscopy (Human Milk Analyzer, Miris, Uppsala, Sweden).

2) Blood biomarkers

IGF-1, IGFBP3, insulin, blood lipids, alpha 1-acid glycoprotein (AGP), high sensitivity C reactive protein (hsCRP) and quantitative amino acids will be measured at the University of Colorado Anschutz Medical Campus CTRC Core lab (30).

We hypothesize that both the gut microbiota and infant growth trajectories are dependent on types of protein-rich foods consumed. We will recruit up to 75 infants/arm and use a conservative 20% attrition which results in 60/group (total completers n = 240). Sample size and power justification are from the longitudinal models with feeding group as the predictor and either gut microbiota (i.e., diversity and Ruminococcus abundance) and growth (i.e., linear growth parameter LAZ, overweight parameter WLZ and growth velocity using the WHO 0–24 months standards) as the outcomes. We completed power simulations (1,000 replicates) using a linear mixed-effects model (LME) with random effects by subject. For the simulations, we used parameter estimates (e.g., fixed and random effects) from our randomized pilot study on the gut microbiota and infant growth trajectories after receiving meat- or dairy-based complementary diet (30). We simulated a variety of scenarios assuming that Plant had microbiota and growth outcomes that were either worse than Dairy, between Dairy and Meat, or the same as Meat. To be conservative, we estimated power after increasing the within-group variability by 50%. The power is above 80% in all simulated scenarios (80–90%) where the effect size was simulated to be at least 50% of that observed in our pilot study (30). The power of most simulation scenarios was below 80% when the effect size was 25% of that observed in the pilot study (30).

Aim 1: Using ANOVA for continuous variables and chi-squared test for categorical variables, we will assess whether matching and randomness held by testing whether potential confounders (e.g., feeding mode, maternal height, maternal BMI, education, sex, race/ethnicity, and smoking during pregnancy) differ between groups. Variables with a p < 0.05 and those that have a strong association with the outcome of interest are included in future models. To model the relationship between diet, time, and diet by time interaction and LAZ and WLZ, we use LME model with random effects by subject. This structure explicitly models multiple time points per infant and can adjust for changes of cofounders during the intervention (e.g., a participant changed from breastfeeding to formula feeding at 9 months). Given a significant diet by time interaction, we will test pairwise differences in growth trajectories using Tukey's multiple comparisons.

Aim 2: We model the effects of diet on the gut microbiota at the end of the intervention and longitudinally, incorporating likely covariates and potential cofounders. 16S amplicon sequencing and metagenomic sequencing will each generate tables of annotated sequence counts. Both datasets will be used to model the diet's impact on microbiota taxonomic composition and functional capacity (i.e., genes and pathways identified through shotgun metagenomics) over the course of the intervention. To evaluate differences in overall taxonomic/functional community composition (i.e., beta-diversity), we use the Microbiota Regression based Kernel Association Test (MiRKAT). MiRKAT models the microbiota using phylogenetic kernels to account for differences in microbial profiles (16S or metagenomics count data) (41). As different measures of dissimilarity are optimal under different conditions, optimal-MiRKAT allows multiple dissimilarity metrics to be tested simultaneously. To model microbial taxonomic and functional profiles over time, we use two methods recently proposed (42). First, we use a bi-exponential distribution to summarize microbiota diversity curves over time. To make inferences, we will incorporate the distribution into a hierarchical model with a random intercept to account for multiple measurements for each person. This approach will model changes in individual taxa, both the rare and common, over time. Additionally, we will evaluate longitudinal changes in common alpha-diversity metrics (e.g., richness, evenness, Shannon diversity, and effective species [e.g., Hill's number (42)]). Finally, we use the control group to build a gut microbiota maturation model via a random forest approach as previously described (24) and compare the intervention groups to this model.

Aim 3: We use LME on growth Z scores and the microbiota over time with random effects by subject, and include group as a covariate. We also complete secondary analyses stratified by group to identify relationships between Z scores and microbiota that differ by group. We will use microbiota diversity measures and relative abundances of individual taxa and genes/pathways to model diversity and composition of the microbiota. We also include necessary confounders, e.g., change from breastfeeding to formula. To evaluate whether taxonomic and/or functional features of the microbiota mediate the relationship between diet and growth, we use a method and R package called MedTest (43). As in classic mediation analysis, MedTest evaluates three models (44). In reference to Figure 2, the first model estimates the direct path between protein-rich foods and the gut microbiota and implements as part of Aim 1 using MiRKAT. The second model estimates the path between protein-rich foods and infant growth and will be implemented using a standard linear model in R. The third model estimates the path between protein-rich foods and growth controlling for the path between gut microbiota and growth and will be implemented using MiRKAT. Using these three models, MedTest tests for significances of the mediation effect of the gut microbiota. Since it is unknown which microbiota features may mediate the relationship, MedTest implements multiple microbiota distances: unweighted (45), weighted (46), and generalized UniFrac (47) as well as Jaccard and Bray-Curtis distances. To ensure a well-controlled type-I error rate, MedTest uses a modified permutation procedure to arrive at an omnibus test over all distances.

We focus on a critical developmental phase that has not been well-studied and propose a novel concept that during early complementary feeding, types of common protein-rich foods will differentially impact infant growth and risk of overweight and that gut microbiota will mediate this impact. To our knowledge, this is the first RCT that directly compared meat, dairy and plant consumptions on infant growth and gut microbiota. We also propose to use non-invasive body composition assessment with doubly labeled water, which also provides a reliable estimation of energy expenditure, a crucial component to assess the risk of overweight. Furthermore, we propose a plausible mechanism that protein-rich foods from different sources could impact infant growth trajectories by modulating the gut microbiota. Overall, this project is innovative because it is a substantively different concept and approach. It will greatly support implementing effective dietary interventions to prevent undesirable infant growth patterns and long-term negative health impact.

Although we will use controlled feeding, we are aware that compliance with the dietary regime is a major challenge. Building on extensive experience, we will utilize the following strategies to optimize and monitor compliance: (1) A weekly multiple-choice log to record food intake based on the monthly recommendations (Table 1) as described in our pilot study (30). (2) Caregivers will be asked to return unconsumed foods/formula at the monthly home visit. (3) Weighed 3-day diet record will be collected at 5, 9, and 12 months; (4) Blood urea nitrogen (BUN), measured at 5 and 12 months is a crude marker of total protein intake, and values of BUN will reflect the total amount of protein consumed for comparisons between groups and over time. We also understand the study procedures may be burdensome to some caregivers. Our team has successfully conducted a number of infant trials with similar or more extensive procedures and low drop-out rate (30, 39, 48, 49). Our experienced team members are dedicated to fully support caregivers and will be available by phone/email and in person, and will make extra home visits if needed.

The study protocol is expected to have minimal risk because there are no invasive procedures except the three blood draws at 5, 12, and 24 months of age, which are necessary to answer the research question. The 12-month blood draw will include analyses (Pb and Hb) routinely obtained for well-child surveillance; results will be provided to the primary care provider to avoid a second clinical blood draw. There is an unlikely risk of food allergies. During screening, the caregiver is asked if the infant has any food restrictions or allergies, including milk/dairy protein allergies (i.e., on a special formula). If the answer is yes, the infant is not eligible to participate. If the participant has no food restrictions or allergies, he/she is unlikely to be allergic to the complementary foods provided because most of the study foods are not considered highly allergenic foods. At the baseline visit, the study coordinator explains the potential risk of allergy development when starting complementary feeding and caregivers are given a list of common signs of food allergies. Food allergy is closely monitored by monthly health questionnaires. If repeated concerns are expressed, the participant will be removed from the study and advised to consult the primary care physician.