Menu
Menu
Menu
Menu
We tested the hypotheses that a more mature or diverse gut microbiota will be positively associated with infant growth and inversely associated with inflammation. We characterized gut microbiota from the stool samples of Malawian infants at 6 mo (n = 527), 12 mo (n = 632) and 18 mo (n = 629) of age. Microbiota diversity and maturity measurements were based on Shannon diversity index and microbiota for age Z-score (MAZ), respectively. Growth was calculated as change in Z-scores for weight-for-age (WAZ), length-for-age (LAZ) and head circumference-for-age (HCZ) from 6 to 12 mo and 12 to 18 mo. Biomarkers of inflammation (alpha-1-acid glycoprotein (AGP) and C-reactive protein (CRP)) were measured at 6 and 18 mo. Multivariable models were used to assess the association of each independent variable with each outcome. Microbiota diversity and maturity were related to growth in weight from 6 to 12 mo, but not to growth in length or head circumference or to growth from 12 to 18 mo. Microbiota diversity and maturity may also be linked to inflammation, but findings were inconsistent.
The data for the study reported in this article were obtained from a clinical trial conducted in Malawi. The details of the study known as the International Lipid-based Nutrient Supplements DYAD (iLiNS-DYAD) trial have been reported previously43,44. Briefly, we enrolled 1391 pregnant mothers above 15 years of age and ≤20 gestational weeks from the antenatal clinics of two health centres and two hospitals in Mangochi district. Based on the sample size needed for the original iLiNS-DYAD trial, 869 mothers were allocated to 18 mo follow-up study after delivery and the singleton infants born to these mothers were participants for the current analysis. Infants born to the remaining 522 women who were assigned to pregnancy intervention only were not included (Fig. 1). At baseline, the following data were collected by trained study personnel: Socio-demographic status, maternal age, height, body mass index (BMI), parity, education, HIV status, hemoglobin concentration, household assets, food security, source of drinking water (tap water vs any other source), access to sanitary facility (water closet or ventilation improved pit latrine vs. none or regular pit latrine) and season. We established maternal HIV status using a whole-blood antibody rapid test (Alere Determine HIV-1/2, Alere Medical Co, Ltd. and Uni-Gold HIV; Trinity Biotech plc)44. Details of the trial have been recorded at the clinical trial registry at the National Institutes of Health (USA) (www.clinicaltrials.gov), under the registration number {“type”:”clinical-trial”,”attrs”:{“text”:”NCT01239693″,”term_id”:”NCT01239693″}}NCT01239693. The trial was conducted by adhering to the Good Clinical Practice guide-lines and ethical standards of the Helsinki Declaration. We obtained ethical clearance for the study from the University of Malawi College of Medicine Research and Ethics Committee (COMREC) and the ethics committee at Tampere University Hospital District, Finland. An informed consent was obtained from each participant before being enrolled into the study. An independent data safety and monitoring board monitored the incidence of suspected serious adverse events (SAEs) during the trial. Mothers were trained to collect faecal samples (generally in the morning) from participating children in their homes at 6, and 12 mo of age. The mothers were provided with sample collection tubes a day before the scheduled sample collection visit. Samples from suspected diarrhoea cases (>3 stools a day and markedly more liquid) were excluded, and the visit was rescheduled for two weeks later. The tubes containing faecal matter were sealed, labelled, and immediately stored in a Ziploc bag on a frozen ice pack in a cooler bag. The samples were transported to a satellite clinic within 6 hours of sample collection for a brief storage at −20 °C before being transported to the central clinic in Mangochi for storage at −80 °C within 48 hours. The samples were later shipped on dry ice to the USA for culture-independent analysis of community composition at Washington University, St. Louis, MO. Sample collection spanned all the three seasons of Malawi: warm-wet season (November-April), cool-dry-winter season (May-August), and hot-dry season (September–October). The isolation of DNA from stool and 16S rRNA gene amplicon sequencing was conducted as described elsewhere45. Stool samples were homogenized by grinding in the presence of liquid nitrogen prior to DNA extraction. DNA libraries were prepared by amplifying the V4 region (~255 bp) of the 16S rRNA gene. The DNA libraries were then sequenced on the Illumina MiSeq platform. Sequence processing and picking of clusters of closely related sequences (operational taxonomic units (OTUs) at 97% sequence identity) were performed in QIIME version 1.9.146. OTU data were filtered using a threshold of at least 0.1% of sequence reads in two or more samples. We employed a Random Forests machine learning model to determine microbiota maturity15,17. The model was generated from an analysis of faecal samples collected from members of a Malawian cohort from birth through the second year of life. The model predicts microbiota age (state of development) based on the abundances of 25 age-discriminatory OTUs17. Microbiota ages of study members predicted by this model were compared to the median microbiota age of chronologically age-matched children in the healthy reference group to generate microbiota-for-age Z-scores (MAZ-scores). The data for the healthy reference group of children were obtained from healthy Malawian children and microbiota maturity was calculated as reported earlier15. Microbiota diversity (based on mean alpha diversity of OTUs in each sample) was measured by calculating the Shannon diversity index using the phyloseq package in R47. Shannon index takes into account both richness and evenness of OTUs in each sample. A larger value indicates higher level of diversity48. For these analyses, Shannon index and MAZ scores were calculated at 6, 12 and 18 mo. Shannon index and MAZ scores were calculated with OTU data that were rarefied to 5000 reads. Growth was assessed as described previously43. The z-score growth variables (WAZ, LAZ, WLZ, HCZ) were calculated by standardizing for age and sex using the WHO Child Growth Standards49. The changes in LAZ, WAZ, WLZ, and HCZ were calculated by taking the difference in z-score between time points, and then dividing by the number of days between the two measurements. The value was then multiplied by the standard number of days for each 6-mo period. Values below −2.0 for WAZ, LAZ, WLZ and HCZ were considered to indicate underweight, stunting, wasting and small head circumference, respectively. To assess inflammation, Alpha-1 acid glycoprotein (AGP) and C-reactive protein (CRP) were analysed from plasma by immunoturbidimetry on the Cobas Integra 400 system auto-analyser (F. Hoffmann-La Roche Ltd, Basel, Switzerland) and reported as g/L and mg/L respectively. High AGP was defined as having AGP value >1.0 g/L and high CRP as having CRP value >5.0 mg/L. All data were analysed using SAS version 9.4 (Cary, NC). Children whose age at a given time point was outside of the pre-specified range for these analyses were excluded. At 6 mo children older than 8 mo were excluded (n = 13), at 12 mo children older than 15 mo were excluded (n = 5) and at 18 mo children older than 21 mo were excluded (n = 2). AGP and CRP were natural log transformed and the β-values and confidence intervals were back-transformed. For continuous outcomes, linear regression (proc glimmix) was used and the β-values and SEs of the predictor are presented. For dichotomous outcomes, logistic regression (proc glimmix) was used and the odds ratios and confidence intervals are presented. These models were fully adjusted for the following pre-specified covariates: intervention group; child age on day of stool collection; maternal age, height, body mass index, parity, education, HIV status, and hemoglobin at enrollment; household assets, food security, source of drinking water (tap water vs any other source), residential location and access to sanitary facility (water closet or ventilation improved pit latrine vs. none or regular pit latrine); season at time of stool sample collection; mode of delivery (vaginal or cesarean); site of delivery; and child sex. The residuals were assessed for outliers and normality. Extreme values were winsorized to the 2.5th and 97.5th percentile, and a sensitivity analysis was conducted with and without winsorized values. The models were inspected for multicollinearity and any covariate that was associated with the predictor with a Spearman’s correlation coefficient greater than 0.7 was removed. To address the hypotheses regarding growth outcomes, we used repeated measures ANCOVA to assess whether Shannon index or MAZ were predictive of change in anthropometric z-scores. The model included two age intervals. For the first age interval, the response variable was the change in anthropometric z-score between 6 and 12 mo and the predictor was MAZ or Shannon index at 6 mo. The anthropometric z-score at 6 mo was included as a covariate to control for status at the beginning of the time interval. For the second age interval, the response variable was the change in anthropometric z-score between 12 and 18 mo, the predictor was MAZ or Shannon index at 12 mo, and the growth measurement z-score at 12 mo was included as a covariate. If an observation for growth or microbiota was missing at one time point, the observation was included for the other time point. The beta-value of MAZ or Shannon index across both intervals was of interest but could not be interpreted from a model that also included the interaction between interval and predictor. Therefore, a separate model was conducted that contained a term for the age interval and a term for the predictor, and the beta-value from this predictor was reported. For these models, we first examined the interaction between time interval and the predictor. If this interaction was significant, it meant that the relationship of the predictor to the outcome differed between the two age intervals, and we then assessed each age interval separately. The β-values of the predictor are presented. These models were fully adjusted and include the covariates previously described. To test the hypothesis that MAZ and Shannon index are related to concurrent inflammation, we performed linear regression in which the predictor was either MAZ or Shannon index at 6 mo, and the outcome was inflammation as measured by log-transformed CRP or log-transformed AGP. Additionally, logistic regression models were performed to assess whether MAZ or Shannon index at 6 mo was associated with high AGP or high CRP. To test the hypothesis that MAZ and Shannon index are related to future inflammation, we repeated the models described above, except that the outcomes were measured at 18 mo and the predictors were measured at 6 or 12 mo. The β-values or odds ratios of the predictor are presented. These models were fully adjusted and include the covariates previously described. For taxa-specific analyses, we used two filters to restrict the OTUs used in analysis. The first filter excluded OTUs that were present in less than 5% of the participants at a given time point. The second filter excluded OTUs with very low abundances of less than 0.1% in a given sample. To identify which taxa were related to growth and inflammation, we performed another set of regression model. We focused this analysis on outcomes for which there was a significant relationship with MAZ or Shannon index. The distributions of the OTU abundances were frequently zero-inflated and non-normal. Therefore, we condensed the OTU abundance data into two separate variables: a) a binary variable for whether the OTU was present (at any abundance), and b) a variable with three categories of OTU abundance (OTU absent, OTU present at a count less than the median, OTU present at a count greater than the median). These models were fully adjusted for all the pre-specified covariates. The Benjamini-Hochberg correction for multiple hypotheses was applied, using a false discovery rate (FDR) of 0.15.
N/A
