Study design and randomization
The WASH Benefits trial was conducted in rural villages in the Gazipur, Mymensingh, Tangail, and Kishoreganj districts of Bangladesh28. The traditional household structure in rural Bangladesh is the compound, where patrilineal families live together and share a common courtyard, and additionally sometimes other resources such as a pond, water source, and latrine. Eight pregnant women who lived near each other were grouped into a study cluster, the unit of randomization, to make it easy for a single community health promoter to walk to each compound. A one km buffer zone around each cluster was enforced in order to prevent spillover from nearby clusters. Eight geographically adjacent clusters formed a block. Using a random number generator, an investigator at UC Berkeley (B.F.A.) block randomized each of the eight geographically adjacent clusters to the double-sized control arm or to one of the six intervention arms in the parent trial: water; sanitation; handwashing; combined water, sanitation, and handwashing (WSH); nutrition, or combined nutrition, water, sanitation, and handwashing (N + WSH). This study only assessed physiological stress, oxidative stress, and DNA methylation in the control and the N + WSH arm of the trial.
Since each intervention delivered had visible physical components (lipid-based nutrient supplement sachets, chlorine tablets and storage vessels, potties, latrines, sani-scoop hoes, and handwashing stations), participants and outcome assessors were not masked. However, laboratory investigators were masked to group assignments and four researchers (A.L., A.N.M., S.T.T., and L.K.), following the pre-registered analysis plan, conducted independent masked statistical analyses. Analyses were replicated once. Results were only unmasked after replication of masked analyses.
Sample size and power calculations
Because the sample size calculations were based on the original environmental enteric dysfunction study49, we assumed that this sample size would be sufficient to assess the stress response and DNA methylation outcomes of this substudy. To estimate the minimum detectable effect of nutrition and WSH interventions on sAA, cortisol, oxidative stress, blood pressure, heart rate, and NR3C1 methylation in the trial, we assumed 135 clusters (average of 5 children per cluster) would be enrolled in this substudy and used the standard deviations in Table 4.
With a range of cluster-level intra-class correlations for repeated measures (0.01 to 0.20), the trial would have 90% power to detect the differences between each intervention arm and the control arm outlined in Table 5.
Ethics
Study protocols were approved by human subjects committees at icddr,b (PR-11063 and PR-14108), the University of California, Berkeley (2011-09-3652 and 2014-07-6561) and Stanford University (25863 and 35583). icddr,b organized a data safety monitoring committee that oversaw the trial. The study protocol is available as a Supplementary Note. Participants provided written informed consent. Because compensation for research participation can be perceived as coercive in low-income settings, instead, families received health information (e.g., blood group testing results) as a token of appreciation.
Participants
Pregnant women in their first or second trimesters and their children were enrolled in the study between 31 May 2012 and 7 July 201328. The trial enrolled pregnant women in the first two trimesters to increase the number of available participants in the study area and to address the inaccuracies of gestational age estimation using self-reported last menstrual period dates. Households that had plans to move in the following year or did not own their own home were excluded in order to minimize loss to follow-up. Households that utilized a water source with high iron were excluded to optimize the effectiveness of the chlorine-based water treatment intervention. The selected study area had low levels of groundwater iron and arsenic, as determined by data from the Department of Public Health Engineering, the British Geological Survey, the Department for International Development National Hydrochemical Survey, and a survey conducted before the study began36. Study staff also conducted surveys where respondents self-reported if there was iron taste in their drinking water or iron staining of their water storage vessels. If the respondent was uncertain about the iron content of their drinking water, study staff used Aquatabs and a digital Hach Pocket Colorimeter II to test the water’s chlorine demand. Households with residual chlorine levels below 0.2 mg/L after 30 min were excluded.
Procedures
The control group did not receive intervention-related household visits. The intervention group received a combination of interventions including a nutrition intervention, a drinking water intervention, a sanitation intervention, and a handwashing intervention; hereafter, this combined intervention group will be referred to as the N + WSH group. Details of the combined intervention in the parent trial were previously described28. Briefly, the nutrition component of the combined intervention consisted of the provision of lipid-based nutrient supplements (LNS; Nutriset, France) that included ≥100% of the recommended daily allowance of 12 vitamins and 9 minerals with 9.6 g of fat and 2.6 g of protein daily for children 6–24 months old and age-appropriate maternal and infant World Health Organization (WHO)/Food and Agriculture Organization (FAO) nutrition recommendations (pregnancy–24 months)36. The drinking water component of the combined intervention included chlorine tablets (Aquatabs; Medentech, Ireland) and safe storage vessels for drinking water. The sanitation component of the combined intervention included child potties, sani-scoop hoes to remove feces, and a double pit latrine for all households. The handwashing component of the combined intervention included handwashing stations with soapy water bottles and detergent soap placed near the latrine and kitchen. To promote behaviors such as treating water, using latrines, and handwashing, local community health promoters visited clusters at least once per week during the first 6 months, and subsequently, at least once every 2 weeks.
One year after intervention, urine samples for oxidative stress analysis were collected in Briggs Pediatric Sterile U-Bags and preserved with 0.1% thimerosal49.
Two years after intervention, we collected saliva specimens. All study activities took place in the children’s homes. In our stress response protocol for cortisol and salivary alpha-amylase measurements, the acute stressor was a venipuncture and caregiver physical separation from the child. Children refrained from ingesting caffeinated products and medicine at least one hour before the venipuncture stress protocol. One hour before the venipuncture, the study team obtained consent and interviewed the caregiver about their child’s medical history. Thirty minutes before the venipuncture, the study team measured maternal and child blood pressure and heart rate. The child’s mouth was rinsed with drinking water 15–20 min prior to the venipuncture. During the period leading up to the venipuncture, the children typically played or slept. Using SalivaBio Children’s Swabs (Salimetrics), three saliva samples were collected during the stress response protocol (5–8 min before stressor onset, 5 min after stressor onset, and 20 min after stressor onset). The stressor was administered at a median time of 10:15 am (IQR: 9:20 am – 11:48 am). For each child, the stressor (the venipuncture and the physical separation of the child from the caregiver) lasted a median of 7.5 min (IQR: 7.5, 7.5). Cortisol was measured at two time points: pre-stressor and 20 min post-stressor. Salivary alpha-amylase was also measured at two time points: pre-stressor and 5 min post-stressor. Additional saliva samples for epigenetic analysis were collected in Oragene kits (OGR-575) and shipped at ambient temperature to EpigenDx (Hopkinton, MA) for DNA methylation analysis of the NR3C1 gene.
At two years, the resting heart rate of participants was measured with a finger pulse oximeter (Nonin 9590 Onyx Vantage) in triplicate, and systolic and diastolic blood pressure were measured with a blood pressure monitor (Omron HBP-1300) in triplicate.
Prespecified outcomes
Analyses were intention-to-treat. We compared the N + WSH arm versus the control arm separately at one year after intervention (median age 14 months) and two years after intervention (median age 28 months). Outcomes included the concentrations of four isomers of F2-isoprostanes [iPF(2α)-III; 2,3-dinor-iPF(2α)-III; iPF(2α)-VI; 8,12-iso-iPF(2α)-VI] measured at one year after intervention. Pre-stressor and post-stressor concentrations of salivary alpha-amylase and salivary cortisol were measured at two years after intervention (median age 28 months). The overall methylation level of the glucocorticoid receptor (NR3C1) exon 1F promoter and the difference in percentage methylation at NGFI-A transcription factor binding site (CpG site 12) in DNA samples were measured at two years. Systolic and diastolic blood pressure and resting heart rate were measured at two years.
Oxidative stress biomarker measurements
F2-isoprostane isomers—iPF(2α)‐III, 2,3‐dinor‐iPF(2α)‐III, iPF(2α)‐VI, and 8,12‐iso‐iPF(2α)‐VI— were quantified by liquid chromatography-tandem mass spectrometry (LC-MS/MS) at Duke University as previously described and optimized for the present study23,50. Urine creatinine (CR) concentration was measured to determine sample volume used for F2-isoprostane analysis. A larger urine volume (300 μL) was used in case of low CR (CR < 0.6 mg/mL; highly diluted urine) to ensure assay sensitivity, a medium volume of urine (200 μL) was used when 0.6 mg/mL <CR < 1 mg/mL, whereas a lower volume (100 μL) was used when CR was high (CR > 1 mg/mL) to decrease the matrix suppression effect on F2-isoprostane signals. To the appropriate volume of urine sample, 20 μL of 1 M HCl, 20 μL of 100 ng/mL internal standard mix [iPF(2α)‐III‐d4, 8,12‐iso‐iPF(2α)‐VI‐d11, iPF(2α)‐VI‐d4], and 1 mL of methyl tert-butyl ether (MTBE) was added and vigorously mixed in FastPrep (Thermo) for 3 × 45 seconds at speed 4. After centrifugation, 800 μL of ether layer was evaporated (nitrogen stream), reconstituted in 50 μL methanol and 70 μL mobile phase A (see below) and 50 μL injected into Shimadzu 20 A series / Applied Biosystems API 4000 QTrap LC/MS/MS instrument. Two C18 columns (Agilent Eclipse Plus, 150 × 4.6 mm and 50 × 4.6 mm, 1.8 µm) in series were used with 0.1% acetic acid as mobile phase A and methanol as mobile phase B delivered as 40–75% B gradient elution over 26 minutes. The mass spectrometer was operated in negative mode with the following MS/MS transitions (m/z): 353/193 [iPF(2α)‐III], 357/197 [iPF(2α)‐III‐d4], 325/237 [2,3‐dinor‐iPF(2α)‐III], 353/115 [iPF(2α)‐VI and 8,12‐iso‐iPF(2α)‐VI], 364/115 [iPF(2α)‐VI‐d11], and 357/115 [8,12‐iso‐iPF(2α)‐VI‐d4]. Lower limits of quantification (LLOQ > 80% accuracy) were 0.063, 0.31, 0.63, and 0.63 mg/mL for iPF(2α)‐III, 2,3‐dinor‐iPF(2α)‐III, iPF(2α)‐VI, and 8,12‐iso‐iPF(2α)‐VI, respectively. The concentration of F2‐isoprostanes was adjusted for urinary creatinine (CR) to account for urine diluteness. Creatinine (CR) was measured after 1/1000 dilution of urine by deionized water, centrifugation, and direct injection into the LC/MS/MS system. Agilent Eclipse Plus 50 × 4.6 mm, 1.8 μm column was used for separation. CR and CR-d3 (internal standard) were measured at m/z = 114/44 and m/z = 117/47, respectively.
Physiological stress and methylation measurements
Pre-stressor and post-stressor salivary alpha-amylase and cortisol were measured following ELISA kit protocols at icddr,b (Salimetrics, Carlsbad, CA). The initial cortisol sample was undiluted, and the initial dilution was 1:200 for salivary alpha-amylase. Out-of-range specimens were rerun at higher or lower dilutions. The coefficient of variation for salivary alpha-amylase and cortisol outcomes was <10%.
Saliva samples that were to be used for the analysis of DNA methylation were collected in Oragene kits (OGR-575) and analyzed by EpigenDx (Hopkinton, MA). EpigenDx performed salivary DNA extraction from Oragene samples, sample bisulfite treatment, PCR amplification, and pyrosequencing and determined percent methylation51. Methylation levels were assessed across the entire glucocorticoid receptor (NR3C1) exon 1F promoter (consisting of 39 assayed CpG sites)52.
First, DNA was extracted from 200 µL saliva using DNAdvance (Beckman Coulter) with the Biomek FXP liquid handler (Beckman Coulter) at EpigenDx (Hopkinton, MA). The NanoDrop 2000 (Thermo Fisher Scientific) was used to quantify the extracted DNA by OD 260/280.
Next, EpigenDx carried out pyrosequencing of bisulfite-treated DNA. Briefly, 500 ng of extracted genomic DNA was bisulfite treated using the EZ DNA Methylation kit (Zymo Research, Inc., CA). The kit protocol was followed for purification and elution of the bisulfite treated DNA (final elution volume of 46 µL). PCR amplification was achieved using 1 µL of bisulfite treated DNA and 0.2 µM of each primer. To purify the final PCR product using sepharose beads, one primer was biotin-labeled and purified by high performance liquid chromatography.
After being bound to Streptavidin Sepharose HP (GE Healthcare Life Sciences), the immobilized PCR products were purified, washed, denatured with a 0.2 µM NaOH solution, and rewashed using the Pyrosequencing Vacuum Prep Tool (Pyrosequencing, Qiagen), according to the manufacturer’s instructions. The NR3C1 pyrosequencing methylation assay target region is listed in Table 6.
Purified single stranded PCR products were annealed to 0.5 µM of sequencing primer. Following the manufacturer’s protocol, 10 µL of the PCR products were pyrosequenced on the PSQ96 HS System (Pyrosequencing, Qiagen). QCpG software (Pyrosequencing, Qiagen) was used to analyze the methylation status of each locus (CpG site) individually as an artificial C/T SNP. To calculate the methylation level at each CpG site, the following formula was used: the percentage of methylated alleles divided by the sum of all methylated and unmethylated alleles. To obtain the mean methylation level, the methylation levels of all measured CpG sites within the targeted region of the gene were used. To ensure detection of incomplete bisulfite conversion of the DNA, each experiment used non-CpG cytosines as internal controls. Other controls in each PCR included unmethylated and methylated DNA. To test for bias, unmethylated control DNA was combined with in vitro methylated DNA at several ratios (0%, 5%, 10%, 25%, 50%, 75%, and 100%), the mixed products were bisulfite-modified and underwent PCR, followed by pyrosequencing analysis.
Statistical analysis
The analysis protocol was pre-registered on Open Science Framework on 14 August 2019. The pre-registered analysis protocol and replication files for the substudy are available (https://osf.io/9573v/). For data management, we used STATA version 14.2. Analyses were conducted using R statistical software version 3.6.1. All biomarker distributions were right-skewed and thus log-transformed. Percentages of methylation were also skewed and therefore logit-transformed.
An individual was classified as a responder to the acute stressor if the difference between pre- and post-stressor cortisol concentrations was a positive change of at least two times the lower limit of sensitivity of the assay (0.014 μg/dL) and a change of at least two times the average coefficient of variation between duplicate tests of the same sample (20%). An individual was classified as a non-responder to the acute stressor if the difference between pre- and post-stressor cortisol concentrations was a negative change with the same thresholds outlined above. An individual was classified as no change if the difference between pre- and post-stressor cortisol concentrations was between these two thresholds, where the difference was not larger than the inherent error in the assay.
We used targeted maximum likelihood estimation with influence curve-based standard errors accounting for clustered observations from the trial’s geographic block-randomized design53.
The randomization of assignment to trial arm resulted in balance in the observed covariates across arms so the primary analysis was unadjusted. For each comparison between arms, we also conducted two secondary adjusted analyses: adjusting for child age and sex assigned at birth only and adjusting for child, age, sex, and covariates found to be significantly related to each outcome (likelihood ratio test P < 0.2). Time of sampling was included as a covariate for salivary alpha-amylase and cortisol analyses only. The full list of covariates is included in the footnotes of the tables.
We conducted a prespecified analysis estimating interactions between child sex and the intervention since biological differences, differential care practices, or other behavioral practices may influence the effect of the N + WSH interventions.
To determine whether missing specimen rates were random, we compared rates of missing specimens across arms and compared characteristics of participants with missing specimens and those with full sets. To account for imbalances in missing outcomes across arms and potential bias due to informative censoring, we repeated the adjusted analysis using inverse probability of censoring weighting, using covariates to predict missing outcomes47,48.
The trial was registered at ClinicalTrials.gov (NCT01590095).
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
