Skip to main content
PMC home page
eLife logoLink to eLife
. 2017 Oct 5;6:e29365. doi: 10.7554/eLife.29365

Effects of water, sanitation, handwashing, and nutritional interventions on telomere length among children in a cluster-randomized controlled trial in rural Bangladesh

Audrie Lin 1,✉, Benjamin F Arnold 1, Andrew N Mertens 1, Jue Lin 2, Jade Benjamin-Chung 1, Shahjahan Ali 3, Alan E Hubbard 1, Christine P Stewart 4, Abul K Shoab 3, Md Ziaur Rahman 3, Md Saheen Hossen 3, Palash Mutsuddi 3, Syeda L Famida 3, Salma Akther 3, Mahbubur Rahman 3, Leanne Unicomb 3, Firdaus S Dhabhar 5, Lia C H Fernald 1, John M Colford Jnr 1, Stephen P Luby 6
Editor: Eduardo Franco7
PMCID: PMC5675593  PMID: 28980942

Abstract

Background:

Shorter childhood telomere length (TL) and more rapid TL attrition are widely regarded as manifestations of stress. However, the potential effects of health interventions on child TL are unknown. We hypothesized that a water, sanitation, handwashing (WSH), and nutritional intervention would slow TL attrition during the first two years of life.

Methods:

In a trial in rural Bangladesh, we randomized geographical clusters of pregnant women into individual water treatment, sanitation, handwashing, nutrition, combined WSH, combined nutrition plus WSH (N + WSH), or control arms. We conducted a substudy enrolling children from the control arm and the N + WSH intervention arm. Participants and outcome assessors were not masked; analyses were masked. Relative TL was measured at 1 and 2 years after intervention, and the change in relative TL was reported. Analysis was intention-to-treat.

Results:

Between May 2012 and July 2013, in the overall trial, we randomized 720 geographical clusters of 5551 pregnant women to a control or an intervention arm. In this substudy, after 1 year of intervention, we assessed a total of 662 children (341 intervention and 321 control) and 713 children after 2 years of intervention (383 intervention and 330 control). Children in the intervention arm had significantly shorter relative TL compared with controls after 1 year of intervention (difference −163 base pairs (bp), p=0.001). Between years 1 and 2, TL increased in the intervention arm (+76 bp) and decreased in the controls (−23 bp) (p=0.050). After 2 years, there was no difference between the arms (p=0.305).

Conclusions:

Our unexpected finding of increased telomere attrition during the first year of life in the intervention group suggests that rapid telomere attrition during this critical period could reflect the improved growth in the intervention group, rather than accumulated stress.

Funding:

Funded by The Bill and Melinda Gates Foundation.

Clinical trial number:

NCT01590095.

Research organism: Human

eLife digest

Stress negatively affects health by causing changes in cells. As a result, excess stress may predispose people to fall ill more often or age faster. It is difficult to measure stress. Some studies suggest that measuring the ends of chromosomes, known as telomeres, may be one way to measure stress. Like the plastic tips on shoelaces, telomeres protect chromosomes from fraying. All peoples’ telomeres shorten over their lifetime with each cell division. Many studies show that telomeres shorten faster in people who experience more stress. When telomeres become too short, cells die faster without being replaced, and the body ages.

Most studies on telomere length have looked at adults. Few studies have looked at children early in life or asked whether there are ways to intervene to stop or reverse stress-related telomere shortening. The first two years of life are a crucial period for the developing brain and immune system, which could set children on a lifelong course toward health or disease. Young children living in low-resource settings often encounter many sources of stress, like poor nutrition, infectious diseases or violence. Studies are needed to determine if interventions in early childhood aimed at reducing some sources of stress improve telomere length or long-term health.

Now, Lin et al. show that interventions to provide safe water, sanitation, handwashing facilities, and better nutrition to children in rural Bangladesh unexpectedly shortened telomeres. As part of a larger study, pregnant women in rural Bangladesh were divided, at random, into groups. One group received a suite of interventions, which included more sanitary toilets, handwashing facilities, and nutritional supplements for their infants. Another group served as a control and did not receive this extra help. Lin et al. looked at telomere length, growth, and infections in a subset of 713 children whose mothers participated in the study.

Children who got the extra help grew faster and were less likely to get diarrhea or parasitic infections than the children in the control group. Unexpectedly, children in the intervention group had shorter telomeres at 14 months of age than the children in the control group. Lin et al. suggest that the telomere shortening in the intervention group might be a consequence of rapid growth and immune system development in the first year of life rather than resulting from biological stress. More studies are needed to ask whether telomere shortening is indeed linked to faster growth and development early in life. The strong and unexpected findings highlight how little is known about how the length of telomeres can be used to predict future health or disease. Interpreting the length of telomeres over a person’s lifetime could prove more nuanced than originally thought.

Introduction

Children in low-income countries often experience infectious diseases and nutritional deficiencies leading to impaired growth, poor development, and early mortality (Black et al., 2017; GBD 2015 Mortality and Causes of Death Collaborators, 2016; Victora et al., 2008). During early life, children exhibit heightened developmental plasticity and are more sensitive to environmental conditions than later in life (Barker, 2007). The theory of developmental origins of health and disease postulates that multiple, cumulative early life exposures to adverse environmental factors may increase allostatic load (the cumulative biological damage from chronic stress) and susceptibility to adult diseases (Barker, 2007; Juster et al., 2010; Price et al., 2013).

Accumulating evidence implicates telomere length (TL) attrition as a potentially important underlying mechanism that links early life insults with adverse health outcomes later in life (Price et al., 2013). Telomeres, the repetitive DNA sequences and protein complexes protecting the ends of linear chromosomes, gradually shorten during normal cell division. Progressive telomere shortening leads to chromosome instability and cell senescence (Blackburn, 2001). Shorter TL has been linked to several age-related conditions including diabetes, heart disease, and early mortality (Cawthon et al., 2003; Fitzpatrick et al., 2007; Salpea et al., 2010). It remains an open debate whether TL serves as a ‘molecular clock’ that gauges cumulative stress exposures over a lifespan or plays a role in the etiology of various diseases (Blackburn et al., 2015; Hamad et al., 2016; Zhan et al., 2015).

Acute or chronic infections may contribute to childhood TL attrition, and inflammation and oxidative stress may be potential mediators (Houben et al., 2008). Infections may induce T-cell proliferation and accelerated telomere attrition (Aviv, 2004). Although no studies have directly investigated the relationship between infection and childhood TL, related studies in adults and animals support the plausibility of an association. Animal models have demonstrated that repeated exposures to Salmonella enterica cause telomere attrition (Ilmonen et al., 2008), and prenatal chronic malaria infections shorten offspring TL (Asghar et al., 2015). In adult humans, Helicobacter pylori infection, hepatitis C virus, HIV, and experimentally induced respiratory infection have been associated with shorter TL (Cohen et al., 2013; Gianesin et al., 2016; Hou et al., 2009; Zanet et al., 2014). Furthermore, caregiver-reported diarrhea in the first two years of life predicted shorter adult TL (Eisenberg et al., 2017). Early life water, sanitation, and handwashing (WSH) interventions could potentially prevent or reduce infections and slow telomere attrition. A systematic review and meta-analysis of WSH interventions reported a reduction in diarrheal illness (pooled estimate of relative risk 0.67, 95% CI 0.59–0.76) (Fewtrell et al., 2005), and a randomized controlled trial in Pakistan showed that promotion of handwashing decreased acute respiratory infections in children (Luby et al., 2005). To our knowledge, no studies have examined the impact of WSH interventions on TL.

Early life nutrition may affect childhood TL. Breast milk could potentially reduce telomere attrition by protecting against inflammation and oxidative stress (Cacho and Lawrence, 2017; Matos et al., 2015) – exposures associated with telomere attrition (Houben et al., 2008). Studies have found an association between exclusive breastfeeding and preschool TL (Wojcicki et al., 2016a), but no association with adult TL (Eisenberg et al., 2017). Improved intake of micronutrients (vitamins and minerals) may promote telomere maintenance (Bull and Fenech, 2008). Studies in adult populations have found mixed evidence for associations between TL, multivitamin usage, and various micronutrients (Cassidy et al., 2010; Liu et al., 2013; Paul et al., 2015; Richards et al., 2007; Xu et al., 2009). To our knowledge, no randomized controlled trials have assessed the effect of nutritional interventions on child TL.

Telomeres shorten fourfold faster in infants compared to adults (Zeichner et al., 1999); however, only a few studies have assessed the potential associations between environmental factors and TL in early childhood, a sensitive window of growth and development (Entringer et al., 2013; Marchetto et al., 2016; Theall et al., 2013a; Theall et al., 2013b; Wojcicki et al., 2016a). The trajectories of infant TL in low-income countries and the potential impact of early life health-improvement interventions on TL are unknown. We conducted a substudy within a randomized trial in rural Bangladesh to evaluate if an intensive, early life nutrition, water, sanitation, and handwashing intervention would slow telomere attrition among children in their first two years of life (Arnold et al., 2013).

Results

The objective of the WASH Benefits trial was to compare the effects of individual and combined interventions on child health in the first two years of life – the critical window to prevent growth faltering (Arnold et al., 2013). Between 31 May 2012 and 7 July 2013, in the overall trial, 5551 compounds (collections of related households) with pregnant women in their first or second trimester were randomly allocated to one of the six intervention groups or a double sized control group as follows: (1) chlorinated drinking water and safe storage vessel, (2) upgraded sanitation (child potties, sani-scoop hoes to remove feces, and a double pit latrine with a hygienic water seal), (3) handwashing promotion (handwashing stations with detergent soap), (4) combined water + sanitation + handwashing (5) nutrition (lipid-based nutrient supplements and age-appropriate recommendations on maternal nutrition and infant feeding practices), (6) combined nutrition + water + sanitation + handwashing (N + WSH), (7) control group, which did not receive any interventions (Figure 1). Community health promoters visited study compounds in the intervention arms to promote behaviors. We conducted a substudy within the trial to evaluate if the N + WSH intervention would slow telomere attrition among children in their first two years of life (Arnold et al., 2013). The substudy team visited 996 index children after 1 year of intervention (Y1) and 1021 children after 2 years of intervention (Y2) in the control and N + WSH arms only. TL outcomes were measured in 66.5% of the children (N = 662) at Y1% and 69.8% of the children (N = 713) at Y2, but not at birth (Figure 1). We expect TL at birth to be similar for the intervention and control arms because household enrollment characteristics were balanced between both arms (Table 1).

Figure 1. Flowchart of participants’ progress through the phases of the trial.

Figure 1.

Open in a new tab
Figure 1—source data 1. Source data and code for Figure 1.

Table 1. Enrollment characteristics within the Control households and the N + WSH intervention households.

Table 1—source data 1. Source data and code for Table 1.
  Children measured at Year 1 Children measured at Year 2
No. of children: Control (N=321) N+WSH Intervention (N=338) Control (N=330) N+WSH Intervention (N=380)
% / mean (SD) % / mean (SD) % / mean (SD) % / mean (SD)
Maternal
 Age (years) 23 (5) 24 (5) 23 (5) 24 (5)
 Years of education 7 (3) 6 (3) 7 (3) 6 (3)
Paternal
 Years of education 5 (4) 5 (4) 6 (4) 5 (4)
 Works in agriculture 24% 29% 25% 29%
Household 
 Number of persons 5 (2) 5 (2) 5 (2) 5 (2)
 Has electricity 60% 62% 62% 62%
 Has a cement floor 16% 12% 15% 13%
 Acres of agricultural land owned 0 (0) 0 (0) 0 (0) 0 (0)
Drinking Water 
 Shallow tubewell primary water source 72% 70% 72% 72%
 Stored water observed at home 49% 53% 49% 52%
 Reported treating water yesterday 0% 0% 0% 0%
 Distance (mins) to primary water source 1 (1) 1 (2) 1 (1) 1 (2)
Sanitation 
 Reported daily open defecation
  Adult men 4% 9% 4% 9%
  Adult women 3% 5% 3% 5%
  Children: 8-<15 years 4% 11% 2% 11%
  Children: 3-<8 years 29% 35% 32% 37%
  Children: 0-<3 years 73% 88% 73% 88%
 Latrine
  Owned 62% 52% 59% 52%
  Concrete slab 97% 93% 96% 94%
  Functional water seal 38% 31% 38% 31%
  Visible stool on slab or floor 54% 48% 52% 46%
 Owned a potty 8% 4% 7% 5%
 Human feces observed in the
  House 6% 7% 5% 9%
  Child’s play area 1% 1% 1% 2%
Handwashing 
 Within 6 steps of latrine
  Has water 18% 13% 21% 14%
  Has soap 9% 6% 11% 7%
 Within 6 steps of kitchen
  Has water 11% 10% 11% 11%
  Has soap 3% 3% 5% 4%
Nutrition 
  *Household is food secure 74% 72% 73% 72%
Open in a new tab

Enrollment characteristics of households with children who had telomere measurements. Data are percentages of binary variables or mean (SD) of continuous variables. Percentages were estimated from slightly smaller denominators than those shown at the top of the table for the following variables due to missing values: mother’s age, father’s education, father works in agriculture, acres of land owned, open defecation, latrine has a concrete slab, latrine has a functional water seal, visible stool on latrine slab or floor, ownership of child potty, observed feces in the house or child’s play area, handwashing variables.

*Assessed by the Household Food Insecurity Access Scale.

In the substudy, a quarter of the fathers were engaged in agriculture. 61% of households reported having electricity available, and 86% had an earthen floor. At enrollment, 72% of households were food secure, 56% of households owned a latrine, and 9% of households had a handwashing station with soap near the latrine. The primary water source for the majority of households (72%) was a shallow tubewell. Respondents reported the occurrence of daily open defecation in 80% of children less than 3 years of age. The substudy household enrollment characteristics were similar to the overall trial (Table 2).

Table 2. Balance of enrollment characteristics in the WASH Benefits main trial, telomere substudy children enrolled at Year 1, and telomere substudy children lost to follow-up at Year 2.

Table 2—source data 1. Source data and code for Table 2.
WASH Benefits Main Trial Telomere substudy: Had telomere outcomes at Year 1 Telomere substudy: Lost to follow-up at Year 2 (from those who had telomere outcomes at Year 1)
No. of children: Control (N=1779) N+WSH Intervention (N=953) Control (N=321) Control (N=338) N+WSH Intervention (N=61) N+WSH Intervention (N=44)
% / mean (SD) % / mean (SD) % / mean (SD) % / mean (SD) % / mean (SD) % / mean (SD)
Maternal
 Age (years) 24 (5) 24 (6) 23 (5) 24 (5) 23 (4) 23 (5)
 Years of education 6 (3) 6 (3) 7 (3) 6 (3) 7 (3) 6 (4)
Paternal
 Years of education 5 (4) 5 (4) 5 (4) 5 (4) 5 (4)  5 (4)
 Works in agriculture 30% 30% 24% 29% 20%  18%
Household 
 Number of persons 5 (2) 5 (2) 5 (2) 5 (2) 5 (3)  5 (2)
 Has electricity 57% 60% 60% 62% 57%  61%
 Has a cement floor 10% 10% 16% 12% 20%  7%
 Acres of agricultural land owned 0.15 (0.21) 0.14 (0.38) 0 (0) 0 (0) 0 (0)  0 (0)
Drinking Water 
 Shallow tubewell primary water source 75% 73% 72% 70% 77%  66%
 Stored water observed at home 48% 48% 49% 53% 56%  57%
 Reported treating water yesterday 0% 0% 0% 0% 0%  0%
 Distance (mins) to primary water source 1 (1) 1 (2) 1 (1) 1 (2) 1 (1)  1 (1)
Sanitation 
 Reported daily open defecation
  Adult men 7% 7% 4% 9% 3%  7%
  Adult women 4% 4% 3% 5% 2%  2%
  Children: 8-<15 years 10% 10% 4% 11% 8%  13%
  Children: 3-<8 years 38% 37% 29% 35% 30%  32%
  Children: 0-<3 years 82% 88% 73% 88% 73%  83%
 Latrine
  Owned 54% 53% 62% 52% 67%  50%
  Concrete slab 95% 94% 97% 93% 100%  98%
  Functional water seal 31% 27% 38% 31% 46%  38%
  Visible stool on slab or floor 48% 46% 54% 48% 64%  53%
 Owned a potty 4% 4% 8% 4% 16%  2%
 Human feces observed in the
  House 8% 7% 6% 7% 10%  5%
  Child’s play area 2% 1% 1% 1% 2%  0%
Handwashing 
 Within 6 steps of latrine
  Has water 14% 11% 18% 13% 18%  8%
  Has soap 7% 6% 9% 6% 11%  8%
 Within 6 steps of kitchen
  Has water 9% 9% 11% 10% 9%  5%
  Has soap 3% 3% 3% 3% 0%  0%
Nutrition 
 *Household is food secure  67%  71%  74%  72%  75%  68%
Open in a new tab

Data are percentages of binary variables or mean (SD) of continuous variables. Percentages were estimated from slightly smaller denominators than those shown at the top of the table for the following variables due to missing values: mother’s age, father’s education, father works in agriculture, acres of land owned, open defecation, latrine has a concrete slab, latrine has a functional water seal, visible stool on latrine slab or floor, ownership of child potty, observed feces in the house or child’s play area, handwashing variables.

*Assessed by the Household Food Insecurity Access Scale.

We measured whole blood relative telomere length by quantitative polymerase chain reaction (qPCR), expressed as the ratio of telomere to single-copy gene abundance (T/S ratio) (Cawthon, 2002; Lin et al., 2010). At the time of TL assessment, the mean (±SD) age was 14.1 (±2.1) months at Y1 and 28.2 (±1.9) months at Y2. TL was normally distributed (Figure 2). The mean (±SD) TL was 1.43 (±0.23) T/S ratio at Y1 (6729 ± 549 base pairs (bp)) and 1.45 (±0.24) T/S ratio at Y2 (6763 ± 586 bp). Although these averages are short for young children compared to some previous findings (Factor-Litvak et al., 2016; Frenck et al., 1998; Wojcicki et al., 2016a), they are similar to a recent report of newborns whose mothers experienced a high level of stress during pregnancy (Marchetto et al., 2016); the stress associated with low socioeconomic status may have contributed to the short average TL in this study.

Figure 2. Kernel density plots summarize the distribution of the telomere lengths (T/S ratios) among enrolled children.

Figure 2.

Open in a new tab

In each panel, a dashed orange line illustrates the distribution of T/S ratio among control children and a solid blue line illustrates the distribution of T/S ratio among N+WSH intervention children. Even if a child was not present at Year 1, they were included in the analysis if they provided a sample at Year 2.

Figure 2—source data 1. Source data and code for Figure 2.

We conducted an intention-to-treat analysis using generalized linear models with robust standard errors. Measures of intervention adherence (including observed hardware availability) were high (over 80%) and sustained throughout Y1 and Y2 (S. P. Luby et al. in review). After 1 year of intervention, compared with controls (mean = 1.47 T/S or 6813 bp), children in the combined intervention arm (mean = 1.40 T/S or 6650 bp) had significantly shorter relative TL (difference −0.07 T/S or −163 bp, p=0.001; Table 3). The entire distribution of the T/S ratio was shifted lower for the intervention arm compared to the control at Y1 (Figure 2). Unadjusted, adjusted, and inverse probability of censoring weighting (IPCW; to adjust for potential bias from loss to follow-up) analyses yielded similar estimates (Table 3). At the Y2 measurement, there was no significant difference (p=0.305) between the intervention arm and the control (Table 3, Figure 2).

Table 3. Effect of intervention on telomere length (T/S ratio) after 1 and 2 years of N + WSH intervention and on change in telomere length between Year 1 and Year 2.

Table 3—source data 1. Source data and code for Table 3.
Arm N Mean Unadjusted difference:
Intervention vs. Control
(95% CI)		 Age- and sex- adjusted
difference: Intervention vs.
Control (95% CI)		 Fully adjusted † difference:
Intervention vs. Control
(95% CI)		 Inverse probability of
censoring † difference: Intervention
vs. Control (95% CI)
After 1 year of intervention (age ~ 14 months)
 Control 321 1.47
 N + WSH 341 1.40 −0.07 (−0.11,–0.03) p=0.001* −0.06 (−0.10,–0.02) p=0.005* −0.05 (−0.10,0.01) p=0.099 −0.07 (−0.11,–0.03) p=0.001*
After 2 years of intervention (age ~ 28 months)
 Control 330 1.46
 N + WSH 383 1.44 −0.02 (−0.06,0.02) p=0.305 −0.02 (−0.06,0.02) p=0.271 −0.04 (−0.09,0.01) p=0.147 −0.02 (−0.06,0.02) p=0.288
Change in telomere length between Year 1 and 2
 Control 260 −0.01
 N + WSH 297 0.03 0.04 (0.00,0.08) p=0.050* 0.04 (−0.00,0.08) p=0.073 0.04 (−0.01,0.09) p=0.081 0.03 (−0.01,0.07) p=0.088
Open in a new tab

*P<0.05.

Confidence intervals were adjusted for clustered observations using robust standard errors. Separate weights were created for the inverse probability weighting for each of the different analyses because the probability of missing at Year 1 was different than at Year 2.

† Adjusted for pre-specified covariates associated with the outcome (likelihood ratio test p-value<0.2): Field staff who collected data, month of measurement, household food insecurity, child age, child sex, mother’s age, mother’s height, mother’s education level, number of children < 18 years in the household, number of individuals living in the compound, distance in minutes to the primary water source, household floor materials, household wall materials, household electricity, and household assets (wardrobe, table, chair, clock, khat, chouki, radio, television, refrigerator, bicycle, motorcycle, sewing machine, mobile phone, cattle, goats, and chickens).

Next, we compared the change in TL between Y1 and Y2. Household enrollment characteristics were balanced between individuals who had TL outcomes at Y1 versus those who were lost to follow-up at Y2 (Table 2). Of index children with a TL measurement at both Y1 and Y2 (N = 557; 54.6% of the total children visited), relative TL increased by +0.03 T/S (+76 bp) in the intervention group and decreased by −0.01 T/S (−23 bp) in the control group (Table 3). Overall, the difference between the control and intervention arms in the change in relative TL from Y1 to Y2 was 0.04 T/S and was borderline significant (p=0.050; Table 3). Unadjusted, adjusted, and inverse probability of censoring weighting analyses generated similar estimates but only the unadjusted model was significant (Table 3).

Finally, we conducted a subgroup analysis by sex because biological differences, differential care practices by sex, or other sex-specific behaviors could modify the effect of the intervention on TL when stratifying by sex (Wojcicki et al., 2016b). Sex was not a significant effect modifier (sex by treatment interaction p=0.435 at Y1 and p=0.105 at Y2; Table 4). Boys had shorter TL than girls at Y1 (−0.07 T/S, p=0.007) and Y2 (−0.09 T/S, p=0.001), consistent with previous studies (Gardner et al., 2014). Although a previous study reported seasonal variation in TL (Rehkopf et al., 2014), we did not observe seasonal changes within this study.

Table 4. Subgroup analysis across sex of the effect of N + WSH intervention on telomere length (T/S ratio) after 1 and 2 years of intervention and on change in telomere length between Year 1 and Year 2.

Table 4—source data 1. Source data and code for Table 4.
Arm Female children Male children Female children Unadjusted difference: Intervention vs. Control (95% CI) Male children Unadjusted difference: Intervention vs. Control (95% CI) Interaction term sex by treatment (95% CI)
N Mean N Mean
After 1 year of intervention (age ~ 14 months)
 Control 162 1.50 159 1.43
 N + WSH 180 1.42 161 1.37 −0.08 (−0.13,–0.03) p=0.001* −0.06 (−0.11,–0.00) p=0.040* 0.03 (−0.04,0.09) p=0.435
After 2 years of intervention (age ~ 28 months)
 Control 167 1.50 163 1.42
 N + WSH 202 1.45 181 1.42 −0.05 (−0.10,0.01) p=0.082 0.00 (−0.05,0.05) p=0.877 0.05 (−0.01,0.12) p=0.105
Change in telomere length between Year 1 and 2
 Control 134 –0.01 126 −0.01
 N + WSH 160 0.03 137 0.04 0.04 (−0.02,0.09) p=0.227 0.05 (0.00,0.09) p=0.048* 0.01 (−0.06,0.08) p=0.747
Open in a new tab

*P<0.05.

Confidence intervals were adjusted for clustered observations using robust standard errors.

Discussion

Here, we demonstrate an effect of an intervention on TL in infants and report the trajectories of infant TL in a low-income country. In rural Bangladesh, an intensive, combined water, sanitation, handwashing, and nutrition intervention delivered to compounds of newborn children increased TL attrition during the critical first year of life. This result is contrary to many studies that reported increased TL attrition associated with prenatal psychosocial stress (Entringer et al., 2013; Marchetto et al., 2016), childhood institutional care (Drury et al., 2012), disease (Fitzpatrick et al., 2007; Salpea et al., 2010), and mortality (Cawthon et al., 2003); however, these prior studies did not examine participants in the same age range (1–2 years) as this study. The difference between the intervention and control arms in the change in relative TL from Y1 to Y2 was borderline significant, and at Y2, we found no difference in relative TL between the arms. In high- and low-resource settings, there is a dearth of evidence on the effect of environmental exposures on TL trajectories in this age range and its implications for adult health outcomes.

The accelerated TL attrition observed during the first year of life in the intervention arm may reflect improved immune system development. The WASH Benefits trial found that children in the intervention arm had reductions in caregiver-reported diarrhea, soil-transmitted helminth infections, and Giardia duodenalis infections compared with the control arm (S. P. Luby et al., A. Ercumen et al., and A. Lin et al. in review). It is plausible that the interventions improved immune system development resulting in subsequent protection against infections or that the decreased exposure to pathogens may have improved immunity. The concept of accelerated TL attrition during the first year of life as a reflection of better immune system development is consistent with studies that have reported rapid telomere attrition and accelerated differentiation of hematopoietic stem cells (HSCs), stem cells that give rise to all lineages of immune cells, during the first year of life. A healthy individual experiences ~ 17 HSC divisions in the first year of life, followed by ~2.5 divisions/year between ages 3–13 years, and ~0.6 divisions/year in adults (Elwood, 2004; Kimura et al., 2010; Rufer et al., 1999; Sidorov et al., 2009). Since TL shortens in proportion to the number of cell replications, HSC progenitor divisions correspond with TL attrition rates (Elwood, 2004; Kimura et al., 2010; Rufer et al., 1999; Sidorov et al., 2009). In the first year of life, children in the control group with poor immune system development would potentially experience less HSC divisions and slower rates of TL attrition compared to the intervention group.

The accelerated TL attrition during the first year of life in the intervention arm may also reflect better linear growth. The intervention improved child linear growth at both Y1 and Y2 measurements (S. P. Luby et al. in review). An evolutionary adaptation of organisms during infections is to restrict growth, and instead redirect nutrients and energy to ensure survival (O'Connor et al., 2008). We postulate that the decreased pathogen exposure and the subsequent reduction in acquired infections are potential mechanisms by which better growth was attained in the intervention arm. Some studies have reported high synchrony between peripheral blood and TL attrition rates within other somatic tissues in the same individual (Daniali et al., 2013; Takubo et al., 2002), while others have found differences (Dlouha et al., 2014; Thomas et al., 2008). Although the results of these prior investigations are equivocal, the accelerated peripheral blood TL attrition in the intervention group during the first year of life could potentially indicate rapid muscle or bone cell division involved in growth. We hypothesize that early life TL may be a proxy measure for growth and development of the immune system, the brain, and other vital tissues. Forthcoming WASH Benefits studies focused on child inflammation and enteric pathogen burden will provide evidence on mechanistic pathways leading to different TL, growth, and development outcomes.

After the initial period of rapid linear growth during Y1, our borderline significant finding that telomere length increased in the intervention arm and decreased in the control arm from Y1 to Y2 seems to represent a preview of the TL trajectories that are set: we hypothesize that the intervention children will experience slower lifetime TL attrition and more physiological resistance to stress-related diseases (Juster et al., 2010). The change in relative TL from Y1 to Y2 was small (0.04 T/S 95% confidence interval 0.00, 0.08) and the p-value borderline (p=0.050). Nevertheless, our diarrhea, soil-transmitted helminth, Giardia duodenalis, and growth results suggest that these interventions are potentially interrupting key infection and malnutrition pathways to reduce exposure to biological adversity and allostatic load (S. P. Luby et al., A. Ercumen et al., and A. Lin et al. in review). In a Filipino cohort, decreased infant diarrheal prevalence was associated with longer adult TL (Eisenberg et al., 2017). In observational studies, various vitamins were associated with longer adult TL (Xu et al., 2009), and early exclusive breastfeeding was associated with longer child TL at ages 4–5 years (Wojcicki et al., 2016a). Our nutrition intervention included promotion of exclusive breastfeeding and micronutrient fortified lipid-based nutrient supplements. The Y1 to Y2 changes in TL that we observed are consistent with these studies. The smaller than expected treatment effect and the borderline p-value indicate that the trial was slightly underpowered to detect differences in the change in relative TL from Y1 to Y2 among the intervention and control groups.

Alternatively, our finding of a modest, borderline impact from Y1 to Y2 may suggest that the effect of the interventions is diminished in children over 1 year of age. This potential waning of intervention effects as a child ages could be due to less consumption of breast milk and thus, reductions in its protective effects or the increased mobility of children in environments contaminated with animal feces. This interpretation would also be consistent with no differences in relative TL observed at Y2 between the two groups, underscoring the importance of targeting interventions early in life during the sensitive period when they are likely to have the largest impact on childhood TL.

This study had several limitations. These results from a rural, low-resource area in Bangladesh may not generalize to other populations. The lack of geographic matching in the substudy could have resulted in imbalances between study arms in factors associated with geography. In addition, differential loss to follow-up in the control and intervention arms may have biased results due to differences in unmeasured characteristics. However, our IPCW analysis showed that bias from loss to follow-up was unlikely based on a large set of measurable characteristics. The lithium heparin additive in the whole blood may have inhibited the qPCR reaction and the long duration of time from sample collection to TL measurement may have affected TL, but any systematic measurement errors would likely affect both study groups equally. These factors may provide potential explanations for the shorter average TL observed in this study compared to other studies (Factor-Litvak et al., 2016; Frenck et al., 1998; Wojcicki et al., 2016a). We only measured TL in whole blood, which might yield different results compared to TL measured in less proliferative tissue types (e.g., muscle or fat) (Daniali et al., 2013). Finally, we did not measure TL at birth (Wojcicki et al., 2016b), but we would expect it to be similar for both groups because household characteristics were balanced by randomization.

Our findings are surprising, and they represent important contributions to the nascent and rapidly expanding field of infant telomere biology. Most studies to date have highlighted shorter TL as a measure of intrauterine or childhood stress (Drury et al., 2012; Entringer et al., 2013; Marchetto et al., 2016; Price et al., 2013), disease (Fitzpatrick et al., 2007; Salpea et al., 2010), and mortality (Cawthon et al., 2003). However, our results motivate an intriguing hypothesis: here, we suggest that during the first year of life, accelerated TL attrition could reflect better child growth, neurodevelopment, and immune function. Although TL was a sensitive outcome that responded to an intervention, this trial underscores our limited understanding of environmental exposures contributing to TL dynamics during early life and the critical physiological pathways linking TL and lifelong health trajectories. Since potential confounding could plague observational studies, evaluating the relationship between modifiable exposures and TL within the context of randomized controlled trials could provide valuable contributions to the field of telomere biology.

Materials and methods

Study design

We conducted a cluster-randomized trial in the rural subdistricts in Gazipur, Mymensingh, Tangail and Kishoreganj districts of Bangladesh. The main trial enrolled geographically matched clusters of compounds that were randomly allocated to a double-sized control or one of the six intervention arms (Arnold et al., 2013). Each compound in rural Bangladesh consists of a collection of households of extended families. Due to logistical constraints for specimen collection, the substudy enrolled a subsample of randomized clusters that was balanced across arms (allocation ratio 1:1), but not geographically matched.

Participants

We enrolled pregnant mothers in their first or second trimester. Exclusion criteria included households with plans to move in the following year, households that did not own their home, and households that drew water from a source with high iron content. The children born to the enrolled pregnant mothers were considered index children and are the focus of this analysis. In this substudy, if any two of the following criteria for moderate to severe dehydration were met, the child was excluded from the venipuncture: (1) restless, irritable, (2) sunken eyes, (3) drinks eagerly, thirsty, (4) pinched skin returns to normal position slowly. Children were also excluded from the venipuncture if they were listless or unable to perform their normal activities. Two children met the exclusion criteria: one child in the control arm was excluded at Year 1, and one child in the intervention arm was excluded at Year 2.

Ethics

Primary caregivers of all children provided written informed consent. The 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). A data safety monitoring committee convened by icddr,b oversaw the study.

Randomization and masking

We formed clusters of 8 neighboring households with eligible pregnant women and created a 1 km buffer around each cluster to prevent spillover between clusters. A block was the equivalent of eight geographically-adjacent clusters. An investigator at UC Berkeley (B.F.A.) used a random number generator to block randomize clusters to one of the six interventions or to a double sized control arm as follows: (1) drinking water treatment and safe storage, (2) sanitation, (3) handwashing, (4) combined water + sanitation + handwashing (WSH) (5) nutrition, (6) combined nutrition + water + sanitation + handwashing (N + WSH) and (7) non-intervention control group. This substudy only included children in the control and the combined N + WSH arms.

Participants and the data collection team were not masked because each intervention delivered had visible hardware. One laboratory investigator (J.L.), who was masked to group assignments, conducted all of the TL measurements. Two investigators (A.L., A.N.M.) conducted independent masked statistical analyses to generate final estimates following the pre-registered analysis protocol. After all masked analyses were replicated, the results were unmasked.

Procedures

The combined N + WSH interventions were previously described (Arnold et al., 2013). Briefly, the components of the combined intervention were as follows: water treatment (Aquatabs; NaDCC) and safe storage vessel, sanitation (child potties, sani-scoop hoes to remove feces, and a double pit latrine with a hygienic water seal), handwashing (handwashing stations near the latrine and kitchen, including soapy water bottles and detergent soap), and nutrition (lipid-based nutrient supplements [Nutriset, Malauny, 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 to 24 months of age and age-appropriate recommendations on maternal nutrition and infant feeding practices). Community health promoters visited study compounds in the intervention arms at least once per week during the first 6 months and at least once biweekly to promote behaviors. The control group received no intervention.

TL was measured at 1 year and 2 years after intervention delivery when the children were approximately ages 14 and 28 months respectively. For each child, trained icddr,b staff collected a 5 ml venipuncture sample from the antecubital area of the arm into a Sarstedt S-monovette lithium heparin collection tube. All specimens were labeled with identification numbers only. Specimens were transported on ice to the laboratory, immediately centrifuged, and whole blood was stored at −80°C. Specimens were shipped on dry ice (−79°C) to Dr. Elizabeth Blackburn’s laboratory at the University of California, San Francisco. The duration of time from sample collection to TL measurement ranged from 8 to 32 months.

Measurement of relative TL

Genomic DNA was extracted from heparin-anti-coagulated whole blood stored at −80°C using QIAamp DNA Mini Kit (QIAGEN, Hilden, Germany). DNA was quantified by measuring OD260 with a NanoDrop 200 c Spectrophotometer (Nanodrop Products, Wilmington, DE). Samples that passed the quality control of OD260/OD280 between 1.7–2.0 were used for TL measurement. Of the 1384 samples, eight did not pass quality control. Of the remaining 1376 samples assayed, one sample was not amplified, resulting in 1375 samples with valid TL data.

TL was measured in whole blood by quantitative PCR (qPCR) using a protocol adapted from the published original method by Cawthon (Cawthon, 2002; Lin et al., 2010). This method determines relative TL (i.e., T/S ratio) by measuring the factor by which each DNA sample differs from a reference DNA sample in its ratio of telomere repeat copy number (T) to single-copy gene copy number (S) (Cawthon, 2002). The primers for the telomere PCR were tel1b [5'-CGGTTT(GTTTGG)5GTT-3'], used at a final concentration of 100 nM, and tel2b [5'-GGCTTG(CCTTAC)5CCT-3'], used at a final concentration of 900 nM. The primers for the single-copy gene (human beta-globin) PCR were hbg1 [5' GCTTCTGACACAACTGTGTTCACTAGC-3'], used at a final concentration of 300 nM, and hbg2 [5'-CACCAACTTCATCCACGTTCACC-3'], used at a final concentration of 700 nM. The final reaction mix contained 20 mM Tris-HCl, pH 8.4; 50 mM KCl; 200 μM each dNTP; 1% DMSO; 0.4x Syber Green I; 22 ng E. coli DNA; 0.4 Units of Platinum Taq DNA polymerase (Invitrogen Inc., Carlsbad, CA), and approximately 6.6 ng of genomic DNA per 11 microliter reaction.

The telomere (T) thermal cycling qPCR profile consisted of denaturing at 96°C for 1 min followed by 30 cycles of denaturing at 96°C for 1 s and annealing or extension at 54°C for 60 s with fluorescence data collection. The single-copy gene (S) thermal cycling qPCR profile consisted of denaturing at 96°C for 1 min followed by 8 cycles of denaturing at 95°C for 15 s, annealing at 58°C for 1 s, and extension at 72°C for 20 s, followed by 35 cycles of denaturing at 96°C for 1 s, annealing at 58°C for 1 s, extension at 72°C for 20 s, and holding at 83°C for 5 s with data collection. The T/S ratio for each sample was measured twice (technical replicates). When the duplicate T/S value and the initial value varied by more than 7%, the sample was run a third time and the two closest values were reported. The average coefficient of variation for TL measurement in this study was 2.1%.

Tubes containing 26, 8.75, 2.9, 0.97, 0.324 and 0.108 ng of a reference DNA (pooled genomic DNA from 100 females) were included in each PCR run so that the quantity of targeted templates in each research sample could be determined relative to the reference DNA sample by the standard curve method. The same reference DNA was used for all PCR runs. To control for inter-assay variability, eight control DNA samples were included in each run. In each batch, the T/S ratio of each control DNA was divided by the average T/S for the same DNA from 10 runs to calculate a normalizing factor. This is done for all eight samples and the average normalizing factor for all eight samples was used to correct the participant DNA samples to calculate the final T/S ratio. The assay was performed as plates of 96 samples. Control and intervention samples were randomly interspersed to minimize potential plate effects. DNA extraction and TL assay were performed in two batches (3.5 months apart). The first batch contained 663 samples and the second batch contained 721 samples. All assays were performed using the same lots of reagents. To adjust for assay batch variations, 48 samples from the first batch were re-assayed together with the second batch of samples and data from the second batch of samples were adjusted based on the systematic difference between the first batch values versus the second batch values for these 48 samples. This adjustment factor was 1.05.

To determine the conversion factor for the calculation of approximate base pair telomere length from T/S ratio, the above method was used to determine the T/S ratios, relative to the same reference DNA, for a set of genomic DNA samples from the ATCC authenticated human fibroblast primary cell line IMR-90 (ATCC: CCL-186, cell line authentication method: STR profiling; ATCC determined the cell line was free of mycoplasma contamination) at different population doublings, as well as with the telomerase protein subunit gene (hTERT) transfected into a lentiviral construct. The mean TRF length from these DNA samples was determined using Southern blot analysis. Comparison of T/S ratios versus base pairs derived from the Southern blot analysis generated the following equation for conversion from T/S ratio to base pairs: base pairs = 3274 + 2413 * (T/S) (Farzaneh-Far et al., 2010).

Outcomes

The pre-specified outcome measures were TL (T/S ratios) measured at 1 year and 2 years after intervention delivery (Y1 and Y2 respectively), and the change in TL from Y1 to Y2.

Statistical analysis

The complete pre-registered analysis protocol is available (https://osf.io/cjjwa/). We provide a summary of our analyses below. Analyses were conducted using R statistical software version 3.2.3 (www.r-project.org).

Sample size

Assuming a standard deviation of 887 base pairs (Entringer et al., 2013), a range of cluster-level intra-class correlations for repeated measures (0.01 to 0.20), and a two-sided alpha of 5%, the substudy had 90% power to detect a 207 to 273 base pair difference between the N + WSH arm and the control.

Statistical parameters

All analyses were intention-to-treat. We compared the N + WSH arm versus the control separately at each time point: 1 year after intervention initiation (Y1; median age 14 months) and 2 years after intervention initiation (Y2; median age 28 months). Between the N + WSH versus control arms, we also compared the change in TL between Y1 and Y2.

The pre-registered analytic approach for these analyses generally followed the same methods as described for the main trial outcomes (Arnold et al., 2013). We used generalized linear models with robust standard errors that account for repeated measures within clusters and reported two-tailed p-values. Randomization led to balance in observed covariates across arms, so, in accordance with our pre-specified analysis plan, we relied on the unadjusted analysis as our primary analysis for the TL outcomes. We conducted two sets of secondary adjusted analyses: (1) adjusted for child age and sex only and (2) fully adjusted for child age, sex, and significantly related covariates (likelihood ratio test p-value<0.20).

We conducted a pre-specified subgroup analysis stratified by sex because biological differences, differential care practices, or other behavioral practices could modify the effect of the combined N + WSH intervention when stratifying by gender.

Missing outcomes

The study carefully tracked enrolled participants and the recovery rates of whole blood specimens per participant. We compared rates of missing blood specimens across randomized arms and also the characteristics of those with missing specimens versus those with a full set of specimens to determine whether missing specimen rates were random. Using the full data (from the main trial) of enrolled mothers in the double-sized control and N + WSH arms, as if every index child with a live birth had TL measured, we used inverse probability weighting to correct for potential bias due to informative censoring (Little et al., 2012). The inverse probability of censoring weighted (IPCW) approach follows best practices for removing potential bias owing to missing outcome measurements in trials, according to a definitive methodologic review conducted by the National Research Council (Little et al., 2012). Logistic regression models were used to predict the probability of TL measurement given baseline characteristics. The baseline characteristics used to fit the models were selected by the same covariate prescreening method used in the fully adjusted analysis. TL differences between the control and intervention group were then estimated using adjusted linear regression weighted by the inverse probability of TL measurement.

Registration

The trial was registered at ClinicalTrials.gov (NCT01590095).

Role of the funding source

The funder approved the study design, but was not involved in data collection, analysis, interpretation or any decisions related to publication. The corresponding author had full access to all study data and final responsibility for the decision to submit for publication.

Data availability

The WASH Benefits data and code that support the findings of this study are available in Open Science Framework (https://osf.io/evc98/).

Acknowledgements

We are indebted to the families who participated in the study and the incredible dedication of the icddr,b staff who delivered the interventions and collected the data and specimens. This study was funded by Global Development grant OPPGD759 from the Bill and Melinda Gates Foundation to the University of California, Berkeley. icddr,b is grateful to the Governments of Bangladesh, Canada, Sweden, and the United Kingdom for providing core/unrestricted support.

Funding Statement

The funder approved the study design, but was not involved in data collection, analysis, interpretation or any decisions related to publication. The corresponding author had full access to all study data and final responsibility for the decision to submit for publication.

Contributor Information

Audrie Lin, Email: audrielin@berkeley.edu.

Eduardo Franco, McGill University, Canada.

Funding Information

This paper was supported by the following grant:

  • Bill and Melinda Gates Foundation Global Development Grant (OPPGD759) to John M Colford.

Additional information

Competing interests

No competing interests declared.

Co-founder of Telomere Diagnostics Inc., a telomere measurement company.

Author contributions

Conceptualization, Data curation, Formal analysis, Funding acquisition, Validation, Investigation, Methodology, Writing—original draft, Project administration, Writing—review and editing.

Resources, Data curation, Software, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Methodology, Project administration, Writing—review and editing.

Resources, Data curation, Software, Formal analysis, Validation, Visualization, Methodology, Writing—review and editing.

Resources, Data curation, Validation, Investigation, Methodology, Writing—review and editing.

Resources, Data curation, Software, Funding acquisition, Validation, Investigation, Project administration, Writing—review and editing.

Data curation, Investigation, Project administration, Writing—review and editing.

Resources, Supervision, Methodology, Project administration, Writing—review and editing.

Supervision, Funding acquisition, Methodology, Project administration, Writing—review and editing.

Data curation, Investigation, Project administration.

Data curation, Investigation, Project administration.

Data curation, Investigation, Project administration.

Data curation, Investigation, Project administration.

Data curation, Investigation, Project administration.

Data curation, Investigation, Project administration.

Resources, Supervision, Funding acquisition, Investigation, Project administration, Writing—review and editing.

Resources, Supervision, Funding acquisition, Project administration, Writing—review and editing.

Funding acquisition, Writing—review and editing.

Funding acquisition, Writing—review and editing.

Resources, Supervision, Funding acquisition, Methodology, Project administration, Writing—review and editing.

Resources, Supervision, Funding acquisition, Methodology, Project administration, Writing—review and editing.

Ethics

Clinical trial registration: The trial was registered at ClinicalTrials.gov (NCT01590095).

Human subjects: Primary caregivers of all children provided written informed consent. The 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).

Additional files

Reporting standard 1
elife-29365-repstand1.docx (132.3KB, docx)
Transparent reporting form

Data availability

The following dataset was generated:

Lin A, Mertens A. 2017. WASH Benefits Bangladesh Analysis of Telomere Outcomes. Publicly available at the Open Science Framework (project no. evc98)

References

  1. Arnold BF, Null C, Luby SP, Unicomb L, Stewart CP, Dewey KG, Ahmed T, Ashraf S, Christensen G, Clasen T, Dentz HN, Fernald LC, Haque R, Hubbard AE, Kariger P, Leontsini E, Lin A, Njenga SM, Pickering AJ, Ram PK, Tofail F, Winch PJ, Colford JM. Cluster-randomised controlled trials of individual and combined water, sanitation, hygiene and nutritional interventions in rural Bangladesh and Kenya: the WASH Benefits study design and rationale. BMJ Open. 2013;3:e003476. doi: 10.1136/bmjopen-2013-003476. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Asghar M, Hasselquist D, Hansson B, Zehtindjiev P, Westerdahl H, Bensch S. Chronic infection. hidden costs of infection: chronic malaria accelerates telomere degradation and senescence in wild birds. Science. 2015;347:436–438. doi: 10.1126/science.1261121. [DOI] [PubMed] [Google Scholar]
  3. Aviv A. Telomeres and human aging: facts and fibs. Science of Aging Knowledge Environment. 2004;2004:pe43. doi: 10.1126/sageke.2004.51.pe43. [DOI] [PubMed] [Google Scholar]
  4. Barker DJ. The origins of the developmental origins theory. Journal of Internal Medicine. 2007;261:412–417. doi: 10.1111/j.1365-2796.2007.01809.x. [DOI] [PubMed] [Google Scholar]
  5. Black MM, Walker SP, Fernald LCH, Andersen CT, DiGirolamo AM, Lu C, McCoy DC, Fink G, Shawar YR, Shiffman J, Devercelli AE, Wodon QT, Vargas-Barón E, Grantham-McGregor S, Lancet Early Childhood Development Series Steering Committee Early childhood development coming of age: science through the life course. The Lancet. 2017;389:77–90. doi: 10.1016/S0140-6736(16)31389-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Blackburn EH. Switching and signaling at the telomere. Cell. 2001;106:661–673. doi: 10.1016/S0092-8674(01)00492-5. [DOI] [PubMed] [Google Scholar]
  7. Blackburn EH, Epel ES, Lin J. Human telomere biology: A contributory and interactive factor in aging, disease risks, and protection. Science. 2015;350:1193–1198. doi: 10.1126/science.aab3389. [DOI] [PubMed] [Google Scholar]
  8. Bull C, Fenech M. Genome-health nutrigenomics and nutrigenetics: nutritional requirements or 'nutriomes' for chromosomal stability and telomere maintenance at the individual level. Proceedings of the Nutrition Society. 2008;67:146–156. doi: 10.1017/S0029665108006988. [DOI] [PubMed] [Google Scholar]
  9. Cacho NT, Lawrence RM. Innate immunity and breast milk. Frontiers in Immunology. 2017;8:584. doi: 10.3389/fimmu.2017.00584. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Cassidy A, De Vivo I, Liu Y, Han J, Prescott J, Hunter DJ, Rimm EB. Associations between diet, lifestyle factors, and telomere length in women. American Journal of Clinical Nutrition. 2010;91:1273–1280. doi: 10.3945/ajcn.2009.28947. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Cawthon RM. Telomere measurement by quantitative PCR. Nucleic Acids Research. 2002;30:e47. doi: 10.1093/nar/30.10.e47. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Cawthon RM, Smith KR, O'Brien E, Sivatchenko A, Kerber RA. Association between telomere length in blood and mortality in people aged 60 years or older. The Lancet. 2003;361:393–395. doi: 10.1016/S0140-6736(03)12384-7. [DOI] [PubMed] [Google Scholar]
  13. Cohen S, Janicki-Deverts D, Turner RB, Casselbrant ML, Li-Korotky HS, Epel ES, Doyle WJ. Association between telomere length and experimentally induced upper respiratory viral infection in healthy adults. JAMA. 2013;309:699–705. doi: 10.1001/jama.2013.613. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Daniali L, Benetos A, Susser E, Kark JD, Labat C, Kimura M, Desai K, Granick M, Aviv A. Telomeres shorten at equivalent rates in somatic tissues of adults. Nature Communications. 2013;4:1597. doi: 10.1038/ncomms2602. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Dlouha D, Maluskova J, Kralova Lesna I, Lanska V, Hubacek JA. Comparison of the relative telomere length measured in leukocytes and eleven different human tissues. Physiological Research. 2014;63:S343–350. doi: 10.33549/physiolres.932856. [DOI] [PubMed] [Google Scholar]
  16. Drury SS, Theall K, Gleason MM, Smyke AT, De Vivo I, Wong JY, Fox NA, Zeanah CH, Nelson CA. Telomere length and early severe social deprivation: linking early adversity and cellular aging. Molecular Psychiatry. 2012;17:719–727. doi: 10.1038/mp.2011.53. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Eisenberg DTA, Borja JB, Hayes MG, Kuzawa CW. Early life infection, but not breastfeeding, predicts adult blood telomere lengths in the Philippines. American Journal of Human Biology. 2017;29:e22962. doi: 10.1002/ajhb.22962. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Elwood N. Telomere biology of human hematopoietic stem cells. Cancer Control. 2004;11:77–85. doi: 10.1177/107327480401100214. [DOI] [PubMed] [Google Scholar]
  19. Entringer S, Epel ES, Lin J, Buss C, Shahbaba B, Blackburn EH, Simhan HN, Wadhwa PD. Maternal psychosocial stress during pregnancy is associated with newborn leukocyte telomere length. American Journal of Obstetrics and Gynecology. 2013;208:134.e1–13134. doi: 10.1016/j.ajog.2012.11.033. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Factor-Litvak P, Susser E, Kezios K, McKeague I, Kark JD, Hoffman M, Kimura M, Wapner R, Aviv A. Leukocyte telomere length in newborns: implications for the role of telomeres in human disease. Pediatrics. 2016;137:e20153927. doi: 10.1542/peds.2015-3927. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Farzaneh-Far R, Lin J, Epel E, Lapham K, Blackburn E, Whooley MA. Telomere length trajectory and its determinants in persons with coronary artery disease: longitudinal findings from the heart and soul study. PLoS One. 2010;5:e8612. doi: 10.1371/journal.pone.0008612. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Fewtrell L, Kaufmann RB, Kay D, Enanoria W, Haller L, Colford JM. Water, sanitation, and hygiene interventions to reduce diarrhoea in less developed countries: a systematic review and meta-analysis. The Lancet Infectious Diseases. 2005;5:42–52. doi: 10.1016/S1473-3099(04)01253-8. [DOI] [PubMed] [Google Scholar]
  23. Fitzpatrick AL, Kronmal RA, Gardner JP, Psaty BM, Jenny NS, Tracy RP, Walston J, Kimura M, Aviv A. Leukocyte telomere length and cardiovascular disease in the cardiovascular health study. American Journal of Epidemiology. 2007;165:14–21. doi: 10.1093/aje/kwj346. [DOI] [PubMed] [Google Scholar]
  24. Frenck RW, Blackburn EH, Shannon KM. The rate of telomere sequence loss in human leukocytes varies with age. PNAS. 1998;95:5607–5610. doi: 10.1073/pnas.95.10.5607. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Gardner M, Bann D, Wiley L, Cooper R, Hardy R, Nitsch D, Martin-Ruiz C, Shiels P, Sayer AA, Barbieri M, Bekaert S, Bischoff C, Brooks-Wilson A, Chen W, Cooper C, Christensen K, De Meyer T, Deary I, Der G, Diez Roux A, Fitzpatrick A, Hajat A, Halaschek-Wiener J, Harris S, Hunt SC, Jagger C, Jeon HS, Kaplan R, Kimura M, Lansdorp P, Li C, Maeda T, Mangino M, Nawrot TS, Nilsson P, Nordfjall K, Paolisso G, Ren F, Riabowol K, Robertson T, Roos G, Staessen JA, Spector T, Tang N, Unryn B, van der Harst P, Woo J, Xing C, Yadegarfar ME, Park JY, Young N, Kuh D, von Zglinicki T, Ben-Shlomo Y, Halcyon study team, Halcyon study team Gender and telomere length: systematic review and meta-analysis. Experimental Gerontology. 2014;51:15–27. doi: 10.1016/j.exger.2013.12.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Gianesin K, Noguera-Julian A, Zanchetta M, Del Bianco P, Petrara MR, Freguja R, Rampon O, Fortuny C, Camós M, Mozzo E, Giaquinto C, De Rossi A. Premature aging and immune senescence in HIV-infected children. AIDS. 2016;30:1363–1373. doi: 10.1097/QAD.0000000000001093. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Hamad R, Walter S, Rehkopf DH. Telomere length and health outcomes: A two-sample genetic instrumental variables analysis. Experimental Gerontology. 2016;82:88–94. doi: 10.1016/j.exger.2016.06.005. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Hou L, Savage SA, Blaser MJ, Perez-Perez G, Hoxha M, Dioni L, Pegoraro V, Dong LM, Zatonski W, Lissowska J, Chow WH, Baccarelli A. Telomere length in peripheral leukocyte DNA and gastric cancer risk. Cancer Epidemiology Biomarkers & Prevention. 2009;18:3103–3109. doi: 10.1158/1055-9965.EPI-09-0347. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Houben JM, Moonen HJ, van Schooten FJ, Hageman GJ. Telomere length assessment: biomarker of chronic oxidative stress? Free Radical Biology and Medicine. 2008;44:235–246. doi: 10.1016/j.freeradbiomed.2007.10.001. [DOI] [PubMed] [Google Scholar]
  30. Ilmonen P, Kotrschal A, Penn DJ. Telomere attrition due to infection. PLoS One. 2008;3:e2143. doi: 10.1371/journal.pone.0002143. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Juster RP, McEwen BS, Lupien SJ. Allostatic load biomarkers of chronic stress and impact on health and cognition. Neuroscience & Biobehavioral Reviews. 2010;35:2–16. doi: 10.1016/j.neubiorev.2009.10.002. [DOI] [PubMed] [Google Scholar]
  32. Kimura M, Gazitt Y, Cao X, Zhao X, Lansdorp PM, Aviv A. Synchrony of telomere length among hematopoietic cells. Experimental Hematology. 2010;38:854–859. doi: 10.1016/j.exphem.2010.06.010. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Lin J, Epel E, Cheon J, Kroenke C, Sinclair E, Bigos M, Wolkowitz O, Mellon S, Blackburn E. Analyses and comparisons of telomerase activity and telomere length in human T and B cells: insights for epidemiology of telomere maintenance. Journal of Immunological Methods. 2010;352:71–80. doi: 10.1016/j.jim.2009.09.012. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Little RJ, D'Agostino R, Cohen ML, Dickersin K, Emerson SS, Farrar JT, Frangakis C, Hogan JW, Molenberghs G, Murphy SA, Neaton JD, Rotnitzky A, Scharfstein D, Shih WJ, Siegel JP, Stern H. The prevention and treatment of missing data in clinical trials. New England Journal of Medicine. 2012;367:1355–1360. doi: 10.1056/NEJMsr1203730. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Liu JJ, Prescott J, Giovannucci E, Hankinson SE, Rosner B, Han J, De Vivo I. Plasma vitamin D biomarkers and leukocyte telomere length. American Journal of Epidemiology. 2013;177:1411–1417. doi: 10.1093/aje/kws435. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Luby SP, Agboatwalla M, Feikin DR, Painter J, Billhimer W, Altaf A, Hoekstra RM. Effect of handwashing on child health: a randomised controlled trial. The Lancet. 2005;366:225–233. doi: 10.1016/S0140-6736(05)66912-7. [DOI] [PubMed] [Google Scholar]
  37. Marchetto NM, Glynn RA, Ferry ML, Ostojic M, Wolff SM, Yao R, Haussmann MF. Prenatal stress and newborn telomere length. American Journal of Obstetrics and Gynecology. 2016;215:94.e1–9494. doi: 10.1016/j.ajog.2016.01.177. [DOI] [PubMed] [Google Scholar]
  38. Matos C, Ribeiro M, Guerra A. Breastfeeding: antioxidative properties of breast milk. Journal of Applied Biomedicine. 2015;13:169–180. doi: 10.1016/j.jab.2015.04.003. [DOI] [Google Scholar]
  39. GBD 2015 Mortality and Causes of Death Collaborators Global, regional, and national life expectancy, all-cause mortality, and cause-specific mortality for 249 causes of death, 1980-2015: a systematic analysis for the Global Burden of Disease Study 2015. Lancet. 2016;388:1459–1544. doi: 10.1016/S0140-6736(16)31012-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. O'Connor JC, McCusker RH, Strle K, Johnson RW, Dantzer R, Kelley KW. Regulation of IGF-I function by proinflammatory cytokines: at the interface of immunology and endocrinology. Cellular Immunology. 2008;252:91–110. doi: 10.1016/j.cellimm.2007.09.010. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Paul L, Jacques PF, Aviv A, Vasan RS, D'Agostino RB, Levy D, Selhub J. High plasma folate is negatively associated with leukocyte telomere length in framingham offspring cohort. European Journal of Nutrition. 2015;54:235–241. doi: 10.1007/s00394-014-0704-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Price LH, Kao HT, Burgers DE, Carpenter LL, Tyrka AR. Telomeres and early-life stress: an overview. Biological Psychiatry. 2013;73:15–23. doi: 10.1016/j.biopsych.2012.06.025. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Rehkopf DH, Dow WH, Rosero-Bixby L, Lin J, Epel ES, Blackburn EH. Seasonal variation of peripheral blood leukocyte telomere length in costa rica: A population-based observational study. American Journal of Human Biology. 2014;26:367–375. doi: 10.1002/ajhb.22529. [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Richards JB, Valdes AM, Gardner JP, Paximadas D, Kimura M, Nessa A, Lu X, Surdulescu GL, Swaminathan R, Spector TD, Aviv A. Higher serum vitamin D concentrations are associated with longer leukocyte telomere length in women. The American Journal of Clinical Nutrition. 2007;86:1420–1425. doi: 10.1093/ajcn/86.5.1420. [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Rufer N, Brümmendorf TH, Kolvraa S, Bischoff C, Christensen K, Wadsworth L, Schulzer M, Lansdorp PM. Telomere fluorescence measurements in granulocytes and T lymphocyte subsets point to a high turnover of hematopoietic stem cells and memory T cells in early childhood. The Journal of Experimental Medicine. 1999;190:157–168. doi: 10.1084/jem.190.2.157. [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Salpea KD, Talmud PJ, Cooper JA, Maubaret CG, Stephens JW, Abelak K, Humphries SE. Association of telomere length with type 2 diabetes, oxidative stress and UCP2 gene variation. Atherosclerosis. 2010;209:42–50. doi: 10.1016/j.atherosclerosis.2009.09.070. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Sidorov I, Kimura M, Yashin A, Aviv A. Leukocyte telomere dynamics and human hematopoietic stem cell kinetics during somatic growth. Experimental Hematology. 2009;37:514–524. doi: 10.1016/j.exphem.2008.11.009. [DOI] [PubMed] [Google Scholar]
  48. Takubo K, Izumiyama-Shimomura N, Honma N, Sawabe M, Arai T, Kato M, Oshimura M, Nakamura K. Telomere lengths are characteristic in each human individual. Experimental Gerontology. 2002;37:523–531. doi: 10.1016/S0531-5565(01)00218-2. [DOI] [PubMed] [Google Scholar]
  49. Theall KP, Brett ZH, Shirtcliff EA, Dunn EC, Drury SS. Neighborhood disorder and telomeres: connecting children's exposure to community level stress and cellular response. Social Science & Medicine. 2013a;85:50–58. doi: 10.1016/j.socscimed.2013.02.030. [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Theall KP, McKasson S, Mabile E, Dunaway LF, Drury SS. Early hits and long-term consequences: tracking the lasting impact of prenatal smoke exposure on telomere length in children. American Journal of Public Health. 2013b;103:S133–S135. doi: 10.2105/AJPH.2012.301208. [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. Thomas P, O' Callaghan NJ, Fenech M. Telomere length in white blood cells, buccal cells and brain tissue and its variation with ageing and Alzheimer's disease. Mechanisms of Ageing and Development. 2008;129:183–190. doi: 10.1016/j.mad.2007.12.004. [DOI] [PubMed] [Google Scholar]
  52. Victora CG, Adair L, Fall C, Hallal PC, Martorell R, Richter L, Sachdev HS, Maternal and Child Undernutrition Study Group Maternal and child undernutrition: consequences for adult health and human capital. The Lancet. 2008;371:340–357. doi: 10.1016/S0140-6736(07)61692-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  53. Wojcicki JM, Heyman MB, Elwan D, Lin J, Blackburn E, Epel E. Early exclusive breastfeeding is associated with longer telomeres in Latino preschool children. The American Journal of Clinical Nutrition. 2016a;104:397–405. doi: 10.3945/ajcn.115.115428. [DOI] [PMC free article] [PubMed] [Google Scholar]
  54. Wojcicki JM, Olveda R, Heyman MB, Elwan D, Lin J, Blackburn E, Epel E. Cord blood telomere length in Latino infants: relation with maternal education and infant sex. Journal of Perinatology. 2016b;36:235–241. doi: 10.1038/jp.2015.178. [DOI] [PubMed] [Google Scholar]
  55. Xu Q, Parks CG, DeRoo LA, Cawthon RM, Sandler DP, Chen H. Multivitamin use and telomere length in women. American Journal of Clinical Nutrition. 2009;89:1857–1863. doi: 10.3945/ajcn.2008.26986. [DOI] [PMC free article] [PubMed] [Google Scholar]
  56. Zanet DL, Thorne A, Singer J, Maan EJ, Sattha B, Le Campion A, Soudeyns H, Pick N, Murray M, Money DM, Côté HC, CIHR Emerging Team Grant on HIV Therapy and Aging: CARMA Association between short leukocyte telomere length and HIV infection in a cohort study: No evidence of a relationship with antiretroviral therapy. Clinical Infectious Diseases. 2014;58:1322–1332. doi: 10.1093/cid/ciu051. [DOI] [PubMed] [Google Scholar]
  57. Zeichner SL, Palumbo P, Feng Y, Xiao X, Gee D, Sleasman J, Goodenow M, Biggar R, Dimitrov D. Rapid telomere shortening in children. Blood. 1999;93:2824–2830. [PubMed] [Google Scholar]
  58. Zhan Y, Song C, Karlsson R, Tillander A, Reynolds CA, Pedersen NL, Hägg S. Telomere length shortening and alzheimer disease--a mendelian randomization study. JAMA Neurology. 2015;72:1202–1203. doi: 10.1001/jamaneurol.2015.1513. [DOI] [PubMed] [Google Scholar]

Decision letter

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Thank you for submitting your article "Effects of an early life intervention on telomere length among children in Bangladesh" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by Eduardo Franco as Reviewing Editor and Prabhat Jha as the Senior Editor.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission. Comments from the reviewers were edited for brevity but the first-person form was kept whenever useful to emphasize a point. Please note that the main points in the compiled critique below are grouped by heading but they may imply corrections or additions at different sections in the revised manuscript.

Summary:

This manuscript presents the results of a randomized controlled trial of a nutrition and sanitation intervention in Bangladesh, examining its effects on telomere length and telomere attrition. The results are intriguing and suggest telomere length may be affected by healthy interventions in the first years of life and lead to improved growth. The central element of the investigation was a relatively large (n=713) longitudinal cohort substudy based on multiple hygiene measures in impoverished households. The authors compared white blood cell (WBC) telomere length (T/S ratio) at 2 time points in young children from control and intervention households. Linear growth of the intervention group improved. Relative to controls, WBC T/S ratio was shorter after 1 year of intervention and similar between groups after 2 years. Between the 2 time points T/S ratio increased in the intervention group and decreased in the controls, a significant difference. In contrast with their hypotheses and other literature, results support an association between rapid growth (reflecting better health) in very early life and telomere shortening.

Essential revisions:

The overarching concerns that should orient the revision are:

1) There should be significant re-framing of the Introduction (i.e., this is a nutrition/sanitation intervention but the authors primarily discuss "adversity").

2) There should be additional analyses to investigate differential loss to follow-up, which is a major source of bias.

3) The results should be reframed as primarily null, emphasizing the more robust models rather than the unadjusted results.

4) More detail must be added to the Materials and methods section regarding blood processing, calculation of telomere length (as absolute base pairs – which are short for newborns/infants, etc.).

5) Discussion on potential mechanisms for the findings, including sources of bias.

Title:

The title as presented is not very descriptive. It should specify that the study is a nutrition and sanitation intervention, as opposed to generically stating "early life" intervention.

Methodological concerns relative to study design:

1) There needs to be more description of the intervention up front, especially since the results may be heavily influenced by the mechanisms through which the intervention is expected to operate. If not in the Results, then at the end of the Introduction. Otherwise it's hard to understand why such an intervention would influence TL and how.

2) It is unclear from the wording whether the control arm is group (1) or an entirely different group that received no intervention. (Results, first paragraph). (In the Materials and methods, a 7th control arm is a listed, which might be a better way to describe it here.)

3) Why weren't the other subjects in the intervention arms included for telomere analysis?

4) Why the focus on adversity? Are there other studies linking TL with nutrition, sanitation, infectious disease, or other exposures more relevant to this intervention?

5) Does this analysis only include children who were interviewed in both Time 1 and Time 2? If so, what percent of the total number of children do these represent? (I.e., how much overlap was there between the 66.5% in wave 1 and the 69.8% in wave 2?)

6) The Introduction suggests that TL may play a causal role in the etiology of disease (Introduction, end of first paragraph), although current evidence does not suggest that this is necessarily the case since most studies are associational. It's important to note that there are questions as to whether TL is simply associated with disease outcomes or actually a cause, particularly in recent Mendelian randomization studies, e.g., Cui et al., Atherosclerosis 2014; Hamad and Rehkopf, Experimental Gerontology 2016; Zhan et al., JAMA Neurology 2015. Perhaps soften this language and add to the discussion of study implications, i.e., we don't really know what the long-term causal implications are for shorter or longer telomeres.

7) Figure 1: Why were new children added to the treatment arm?

8) It's not clear how long the intervention operated, e.g., how long health promoters visited the compounds, whether the structural changes continued to be available throughout. I know that there is a citation to another paper that describes it in more detail, but we need a bit more detail here to understand why this intervention might influence TL.

Methodological concerns relative to measurement of telomere length:

1) More information and data should be included on how the telomere length (in bps) was calculated. 6763 bp is short for infants.

2) Has it been validated to wait so long from sample collection to measurement of TL (subsection “Procedures”, last paragraph)? If not, this is a limitation.

3) Were there differences in the month in which the treatment and control groups were measured? If so, there are known seasonal differences in TL at leukocytes at least (Rehkopf et al., AJHB 2014). Please discuss.

4) The blood sample for telomere length (by PCR) was stated to be lithium heparin, yet heparin inhibits PCR reactions. Sometimes this can be overcome by use of heparinise. Can the authors please clarify the type of specimen used, and if it was Li-heparin how this may have impacted the key results.

5) How measuring TL in leucocytes might influence the results, if the proposed mechanism of the intervention is largely infectious or nutritional? This should be listed as a limitation also, since TL measured in different tissues might yield different results based on a given intervention's mechanisms.

Concerns related to analysis and interpretation:

1) A critical issue is that unadjusted results are presented as the primary, and it is stated that the adjusted and IPCW results are "similar" (Results, fourth paragraph). This is not actually the case. While the point estimates are similar, they are no longer statistically significant, for the most part, in the adjusted and IPCW models. It would be more appropriate to conclude that this study showed less strong effects of the intervention on TL, than the authors imply. It seems inappropriate to consider the unadjusted analysis as primary, since so much differential loss to follow-up occurred.

2) Table 1 does not present the results for whether baseline covariates were statistically significantly different from one another for the two groups. Were there differences in sociodemographic characteristics between the two groups at Y1 and Y2? I.e., differential loss to follow-up that might explain the differences in TL and attrition? This is presented in a table but should be discussed here.

3) Many more children withdrew from the control group. This again highlights the need to present study characteristics to examine possible differential loss to follow-up. And this should be included as a major limitation, in that unmeasured characteristics may be driving these differences and the subsequent results. It also suggests that perhaps the adjusted analyses should be the primary ones.

4) Again, it would be more appropriate to frame this as a study that found null effects except for Year 1 differences, and to discuss why there may have been no effect observed. Moreover, there is not enough discussion of the hypothesized pathways linking adversity and TL, nor on how this intervention addresses adversity as opposed to nutrition and sanitation, which are very different exposures.

5) The authors should postulate mechanisms by which telomere shortening improves growth.

6) If telomere length shortening is associated with better immune functioning and the infants had improved hygienic conditions, then it is counterintuitive when considered in context with germ theory of disease. This should be amended and discussed.

7) The authors discuss adversity, which conjures to mind poverty, psychosocial stress, etc. It is inappropriate to talk about adversity because it was not measured via socioeconomic or psychosocial risk factors, so much as nutrition and sanitation. Please clarify what "adversity" means in these prior studies, and how/whether that relates to this particular intervention.

8) The limitations section needs to be expanded, e.g., lack of process evaluation measures that might clarify mechanistic pathways linking intervention to disease; limited understanding of how different types of exposures might influence TL; differential loss-to-follow-up on unmeasured characteristics that can't be accounted for by IPCW; capturing TL only in a single tissue. What are the implications of the lack of geographic matching for the substudy? What are the implications for generalizability and bias of the exclusion criteria, which eliminated potentially more disadvantaged individuals and therefore may have selected on better outcomes?

9) Only including "significantly related covariates" is not an appropriate way to select covariates. You should include all of those that are theoretically potential confounders. I.e., covariates should be selected conceptually and not empirically. It doesn't seem necessary to include (1) just age child age and sex; just do one fully adjusted model with all relevant covariates (subsection “Statistical parameters”, second paragraph).

10) The authors state that they conducted inverse probability weighting, but do not give these details (subsection “Missing outcomes”).

11) For both Tables 1 and 2, it's confusing to have mean (SD) and N (%) in the same column, and the tables overall are clutters with all the "%" signs. Perhaps have mean (SD) for continuous outcomes and just the% for binary outcomes?

12) Table 1: Is this table for the children who had telomere outcomes measured? Please specify in the footnote.

13) Table 1: At the top left, the table says "No. of compounds." Are the "N =" at the top of the second and third columns therefore number of compounds or number of children? If number of compounds, please present number of children also/instead. Similarly for the top of Table 2.

14) Tables 3 and 4 are very redundant. Can't you just add the IPCW column from Table 4 to Table 2? Perhaps put the CI below the point estimate so it can all fit?

15) The footnotes of Tables 3 and 4 say that covariates were pre-specified, but the text says that adjustment included significantly related covariates at p < 0.2. Please clarify.

16) Table 5 would look better if girls and boys were listed side by side instead of all in the same column. And actually, presenting it stratified in this way doesn't give us the information we want, which is the point estimate and CI for the interaction term of gender x treatment group. Are the headings for the mean and N columns reversed?

17) It is not intuitive why the argument for TL measurement is strengthened as a "potential target for future intervention development and evaluation." To the contrary, this suggests that the simple belief that longer TL = healthier needs to be further nuanced, and that additional studies like this one need to be done. We need a better understanding of the human physiology linking different types of socioeconomic and biomedical exposures with different types of telomere-related outcomes, since the existing literature suffers from too much confounding to understand what is actually going on. I'd say the main takeaway from this study is that we have a very limited understanding of how TL in humans actually responds to different types of interventions, and limited understanding of how TL actually influences different aspects of human health.

18) The presentation of the results in the Abstract is confusing, and doesn't parallel the presentation in the main manuscript. It would be helpful if the authors stated (1) the difference in TL between treatment and control at year 1; (2) the difference in TL between treatment and control at year 2; and (3) the difference in telomere attrition between treatment and control at year 1 and at year 2. These are the primary outcomes, as I understand them.

19) Unclear wording: does linear growth refer to telomeres or the child's height/length? (Abstract) If the latter, then this was not examined in this manuscript and should not go in the Abstract.

eLife. 2017 Oct 5;6:e29365. doi: 10.7554/eLife.29365.sa2

Author response

Essential revisions:

The overarching concerns that should orient the revision are:

1) There should be significant re-framing of the Introduction (i.e., this is a nutrition/sanitation intervention but the authors primarily discuss "adversity").

Thank you for this helpful recommendation. We have significantly revised and re-framed the Introduction section. We also added two new paragraphs. One paragraph focuses on the mechanisms by which the water, sanitation, and handwashing intervention could potentially affect TL: “Acute or chronic infections may contribute to childhood TL attrition, and inflammation and oxidative stress may be potential mediators (Houben, Moonen, van Schooten, and Hageman, 2008). […] To our knowledge, no studies have examined the impact of WSH interventions on TL.”

The other paragraph focuses on the mechanisms by which the nutritional intervention could potentially affect TL: “Early life nutrition may affect childhood TL. […] To our knowledge, no randomized controlled trials have assessed the effect of nutritional interventions on child TL.”

2) There should be additional analyses to investigate differential loss to follow-up, which is a major source of bias.

We agree with the reviewers that differential loss to follow-up could be a source of bias in this trial. This is an issue we had anticipated from the trial’s conception and included in our pre-specified and pre-registered analysis plans for the trial (all outcomes, not just telomere outcomes). In our original submission to eLife we included results from an inverse probability of censoring weighted (IPCW) analysis, which re-weights the analysis population to reflect the enrollment population using measured characteristics. The IPCW approach follows best practices for removing potential bias owing to missing outcome measurements in trials, according to a definitive methodologic review conducted by the National Research Council (Little et al., 2012). In this substudy, the IPCW analysis led to very similar inference as the adjusted and unadjusted analyses (original submission Table 4, revised submission Table 3), which suggests that outcomes were missing at random with respect to measurable characteristics and the missing outcomes are not a likely source of bias in the study. We made the following revisions to help emphasize the IPCW methods and results, and to help readers more clearly evaluate and interpret this sensitivity analysis in light of our primary findings:

-To address whether there were differences in enrollment characteristics between those with TL outcomes at Y1 versus Y2, we revised Table 1 to include treatment group means by year of measurement.

-As recommended by the reviewers, we combined the original Tables 3 and 4 into one table by adding the IPCW column from Table 4 to Table 3. This change enables readers to more clearly interpret the IPCW analysis beside the unadjusted and adjusted models.

-We have added more details on IPCW analysis to the Methods section: “The inverse probability of censoring weighted (IPCW) approach follows best practices for removing potential bias owing to missing outcome measurements in trials, according to a definitive methodologic review conducted by the National Research Council (Little et al., 2012). […] TL differences between the control and intervention group were then estimated using adjusted linear regression weighted by the inverse probability of TL measurement.”

-In the Discussion section, we added this language to the limitations paragraph: “Differential loss to follow-up in the control and intervention arms could potentially introduce bias if differential loss to follow-up was caused by unmeasured characteristics. However, our IPCW analysis showed that bias from loss to follow-up was unlikely based on a large set of measurable characteristics.”

3) The results should be reframed as primarily null, emphasizing the more robust models rather than the unadjusted results.

We agree with the reviewers and have reframed the Results and Discussion accordingly.

Typically, in a clinical trial, adjusted analyses (including IPCW) are pursued for two main reasons: 1) to remove bias from chance imbalances in group characteristics despite randomization or differential loss to follow-up, and 2) to increase the precision of the effect estimate (Pocock, Assmann, Enos, and Kasten, 2002). Balancing these objectives is a classic bias-variance tradeoff in statistics, and our approach, pre-specified in our analysis plan (https://osf.io/cjjwa/), has been to focus first on the issue of bias: namely, do we see evidence for any change in effect after adjustment? Then, if we see no evidence for bias, focus on the unadjusted analysis for our inference. This emphasis follows guidelines from leading statisticians in clinical trials, for example:

“Experiments offer more reliable evidence on causation than observational studies, which is not to gainsay the contribution to knowledge from observation. Experiments should be analyzed as experiments, not as observational studies. A simple comparison of rates might be just the right tool, with little value added by ‘sophisticated’ models.”

Freedman DA. Statistical models for causation – What inferential leverage do they provide? Eval Rev. 2006;30: 691–713.

“Experience shows that for most clinical trials, analyses which adjust for baseline covariates are in close agreement with the simpler unadjusted treatment comparisons. This is because (a) the randomization usually results in well balanced treatment groups, and (b) most covariates are not strongly related to the outcome.”

Pocock SJ, Assmann SE, Enos LE, Kasten LE. Subgroup analysis, covariate adjustment and baseline comparisons in clinical trial reporting: current practice and problems. Stat Med. 2002;21: 2917–2930.

As the reviewer notes below in comment #1 “Concerns related to analysis and interpretation”, the point estimates across analyses are very similar, which we find reassuring because it is consistent with good balance from randomization and outcome measurements missing at random (with respect to measurable characteristics). As the reviewer also notes, the confidence intervals widen slightly (and thus P-values increase slightly) as we move from the unadjusted analysis to the adjusted analyses. An analysis will only gain precision if all covariates are very strongly associated with the outcome (Pocock et al. 2002). Although our pre-specified algorithm pre-screened covariates that were associated with the outcome (a necessary condition for confounding), the adjusted analyses still lost a small amount of precision – suggesting that our conservative inclusion to reduce all possible bias led to a small loss in precision in this study. Given the similarity of the point estimates across analyses (Table 3), we feel that the unadjusted results, which rely only on randomization, are the most parsimonious summary of the trial.

We summarized our rationale in the Materials and methods section of our original submission: “Randomization led to balance in observed covariates across arms, so, in accordance with our pre-specified analysis plan, we relied on the unadjusted analysis as our primary analysis for the TL outcomes.”Tables 1 and 2 provide evidence that the groups with Y1 and Y2 TL outcomes were balanced across enrollment characteristics and that the group lost to follow up did not differ in enrollment characteristics from those with outcomes.

In the Results section, we addressed the reviewer’s concerns regarding the specific statement about unadjusted, adjusted, and IPCW analyses being “similar”. We clarified this statement for the change in TL from Y1 to Y2 result by emphasizing which model was significant: “Unadjusted, adjusted, and inverse probability of censoring weighting analyses generated similar estimates but only the unadjusted model was significant (Table 3).”Although we stated that the difference between the control and intervention arms in the change in relative TL from Y1 to Y2 was borderline significant (in the Results section), we agree that this borderline significant result is overemphasized in the Discussion section. Therefore, in the Discussion section, we have reframed this study as a study that found significant results at Y1, borderline significant/null effects in the difference in the relative change in TL between Y1 and Y2 among the two groups, and null effects at Y2.

To emphasize this point, we have added the borderline significant effects in the difference in the relative change in TL between Y1 and Y2 among the two groups and the null effects at Y2 to the first paragraph of the Discussion section: “The difference between the intervention and control arms in the change in relative TL from Y1 to Y2 was borderline significant, and at Y2, we found no difference in relative TL between the arms.”

In accordance with the guidelines for the interpretation of borderline significant results in clinical trials (Hackshaw and Kirkwood, 2011), we have softened the language of our interpretation and put forth alternative explanations.

The revised language in the Discussion emphasizes the borderline aspect of the change in TL result and enables the reader to interpret whether or not the difference is biologically meaningful and/or if the mechanisms proposed are plausible: “After the initial period of rapid linear growth during Y1, our borderline significant finding that telomere length increased in the intervention arm and decreased in the control arm from Y1 to Y2 seems to represent a preview of the TL trajectories that are set: we hypothesize that the intervention children will experience slower lifetime TL attrition and more physiological resistance to stress-related diseases (Juster, McEwen, and Lupien, 2010). […] The smaller than expected treatment effect and the borderline p-value indicate that the trial was slightly underpowered to detect differences in the change in relative TL from Y1 to Y2 among the intervention and control groups.”

Additionally, we added a new paragraph with an alternative interpretation if the reader considers the borderline result to be null: “Alternatively, our finding of a modest, borderline impact from Y1 to Y2 may suggest that the effect of the interventions is diminished in children over 1 year of age. […] This interpretation would also be consistent with no differences in relative TL observed at Y2 between the two groups, underscoring the importance of targeting interventions early in life during the sensitive period when they are likely to have the largest impact on childhood TL.”

4) More detail must be added to the Materials and methods section regarding blood processing, calculation of telomere length (as absolute base pairs – which are short for newborns/infants, etc.).

We added a paragraph to the Materials and methods section with these details (below).

The new paragraph in the Materials and methods section reads: “To determine the conversion factor for the calculation of approximate base pair telomere length from T/S ratio, the above method was used to determine the T/S ratios, relative to the same reference DNA, for a set of genomic DNA samples from the ATCC authenticated human fibroblast primary cell line IMR-90 (ATCC: CCL-186, cell line authentication method: STR profiling; ATCC determined the cell line was free of mycoplasma contamination) at different population doublings, as well as with the telomerase protein subunit gene (hTERT) transfected into a lentiviral construct. […] Comparison of T/S ratios versus base pairs derived from the Southern blot analysis generated the following equation for conversion from T/S ratio to base pairs: base pairs = 3274 + 2413 * (T/S) (Farzaneh-Far et al., 2010).”

We agree that the averages of 6729 bp at Year 1 and 6763 bp at Year 2 are short for newborns compared to some previous findings (Factor-Litvak et al., 2016; Frenck, Blackburn, and Shannon, 1998; Wojcicki, Heyman, et al., 2016), but these averages are similar to a recent report of newborns whose mothers experienced a high level of stress during pregnancy (Marchetto et al., 2016). It is possible that the short telomere length in our cohort is due to the overall low socioeconomic status, which is shown to be associated with shorter telomere length, although we cannot rule out other factors that may contribute to this, such as race, malnutrition, or infection status.

We have added new language to the Results section: “Although these averages are short for young children compared to some previous findings (Factor-Litvak et al., 2016; Frenck, Blackburn, and Shannon, 1998; Wojcicki, Heyman, et al., 2016), they are similar to a recent report of newborns whose mothers experienced a high level of stress during pregnancy (Marchetto et al., 2016); the stress associated with low socioeconomic status may have contributed to the short average TL in this study.”

5) Discussion on potential mechanisms for the findings, including sources of bias.

We have expanded the discussion on potential mechanisms for the findings by adding a new paragraph to the Discussion section. Please see our response to comment #5 under “Concerns related to analysis and interpretation” for additional details (below).

The new paragraph reads: “The accelerated TL attrition during the first year of life in the intervention arm may also reflect better linear growth. […] Forthcoming WASH Benefits studies focused on child inflammation and enteric pathogen burden will provide evidence on mechanistic pathways leading to different TL, growth, and development outcomes.”

We also vastly expanded the limitations paragraph on sources of bias in the Discussion section. Please see our response to comment #8 under “Concerns related to analysis and interpretation” for additional details (below).

The revised limitations paragraph reads: “This study had several limitations. […] Finally, we did not measure TL at birth (Wojcicki, Olveda, et al., 2016), but we would expect it to be similar for both groups because household characteristics were balanced by randomization.”

Title:

The title as presented is not very descriptive. It should specify that the study is a nutrition and sanitation intervention, as opposed to generically stating "early life" intervention.

The eLife guidelines require titles to be a maximum of 120 characters; therefore, we generated our original title to meet these guidelines.

As suggested, we have changed the title to “Effects of water, sanitation, handwashing, and nutritional interventions on telomere length among children in a cluster-randomized controlled trial in rural Bangladesh.” These changes result in a 147-character title, slightly exceeding the maximum character limit. We included “cluster-randomized controlled trial” in the title according to CONSORT guidelines for randomized controlled trials.

Methodological concerns relative to study design:

1) There needs to be more description of the intervention up front, especially since the results may be heavily influenced by the mechanisms through which the intervention is expected to operate. If not in the Results, then at the end of the Introduction. Otherwise it's hard to understand why such an intervention would influence TL and how.

We added details of the intervention hardware to the description of the intervention in the Results section. Further details are included in the original Materials and methods section.

The revised sentence reads: “Between 31 May 2012 and 7 July 2013, in the overall trial, 5551 compounds (collections of related households) with pregnant women in their first or second trimester were randomly allocated to one of the six intervention groups or a double sized control group as follows: 1) chlorinated drinking water and safe storage vessel, 2) upgraded sanitation (child potties, sani-scoop hoes to remove feces, and a double pit latrine with a hygienic water seal), 3) handwashing promotion (handwashing stations with detergent soap), 4) combined water + sanitation + handwashing 5) nutrition (lipid-based nutrient supplements and age-appropriate recommendations on maternal nutrition and infant feeding practices), 6) combined nutrition + water + sanitation + handwashing (N+WSH), 7) control group, which did not receive any interventions (Figure 1). Community health promoters visited study compounds in the intervention arms to promote behaviors.”

2) It is unclear from the wording whether the control arm is group (1) or an entirely different group that received no intervention. (Results, first paragraph). (In the Materials and methods, a 7th control arm is a listed, which might be a better way to describe it here.)

The control arm received no intervention – we have clarified this in the revision following the reviewer’s suggestion. Please see our response to comment #1 under “Methodological concerns relative to study design” for additional details (above).

3) Why weren't the other subjects in the intervention arms included for telomere analysis?

Venipuncture samples were only collected from children in the control arm, the nutrition arm, the combined water + sanitation + handwashing (WSH) arm, and the combined N+WSH arm. Originally, we proposed telomere analysis in all four of these arms. However, we were not able to obtain sufficient funding to include the other intervention arms (nutrition arm and combined WSH arm) in the telomere analysis. After the publication of these results, we hope to obtain additional funding for a future study on the banked biological specimens that includes the other intervention arms (the individual nutrition arm and the combined WSH arm). This future study would determine if the observed differences in telomere length at Year 1 between the control and combined N+WSH arms were due to nutrition alone, combined WSH only, or the combination of N+WSH.

4) Why the focus on adversity? Are there other studies linking TL with nutrition, sanitation, infectious disease, or other exposures more relevant to this intervention?

We have added two new paragraphs to the Introduction section (please refer to our response to comment #1 in ‘Essential revisions’ above for details) focused on background literature related to the interventions. The new paragraphs include adult studies focused on infections, infectious diseases, and multivitamin supplementation. We originally focused on adversity because the few studies conducted in children measured TL in the context of adversity. To date, the studies linking TL with nutrition and infectious disease were mostly focused on adults in high-income countries. We do not know of any studies linking TL with handwashing, drinking water, or sanitation interventions.

5) Does this analysis only include children who were interviewed in both Time 1 and Time 2? If so, what percent of the total number of children do these represent? (I.e., how much overlap was there between the 66.5% in wave 1 and the 69.8% in wave 2?)

It is unclear to which analysis this comment pertains to. Overall, the analysis at Year 1 includes children with TL measurements at Year 1 (Time 1). The 66.5% is the percentage of children with TL outcomes at Year 1 of the total number of children visited at Year 1 (662/996=66.5%). The analysis at Year 2 includes children with TL measurements at Y2 (Time 2). The 69.8% is the percentage of children with TL outcomes at Year 2 of the total number of children visited at Year 2 (713/1021=69.8%). The analysis of the change in TL from Year 1 to Year 2 only included children with TL outcomes at both time points (N=557). These 557 children represent 54.6% of the total children visited between Year 1 and Year 2 (557/1021). The number of children included in each analysis is displayed in Table 3.

We have added this percentage to the following sentence in the Results section: “Of index children with a TL measurement at both Y1 and Y2 (N=557; 54.6% of the total children visited), relative TL increased by +0.03 T/S (+76 bp) in the intervention group and decreased by -0.01 T/S (-23 bp) in the control group (Table 3).”

6) The Introduction suggests that TL may play a causal role in the etiology of disease (Introduction, end of first paragraph), although current evidence does not suggest that this is necessarily the case since most studies are associational. It's important to note that there are questions as to whether TL is simply associated with disease outcomes or actually a cause, particularly in recent Mendelian randomization studies, e.g., Cui et al., Atherosclerosis 2014; Hamad and Rehkopf, Experimental Gerontology 2016; Zhan et al., JAMA Neurology 2015. Perhaps soften this language and add to the discussion of study implications, i.e., we don't really know what the long-term causal implications are for shorter or longer telomeres.

Thank you for the recommendations and for the references. We completely agree with the reviewer’s comment about uncertainty as to the etiologic and causal relationship between telomeres and disease, and have followed the suggestion to soften the language of this sentence – also adding two of the suggested references (Hamad, Walter, and Rehkopf, Experimental Gerontology 2016 and Zhan et al., JAMA Neurology 2015).

The revised sentence reads:“It remains an open debate whether TL serves as a “molecular clock” that gauges cumulative stress exposures over a lifespan or plays a role in the etiology of various diseases (Blackburn, Epel, and Lin, 2015; Hamad, Walter, and Rehkopf, 2016; Zhan et al., 2015).”

We have revised this sentence in the Discussion section: “Future studies are needed to understand the environmental exposures contributing to TL dynamics during early life and the critical physiological pathways linking TL and lifelong health trajectories.”

7) Figure 1: Why were new children added to the treatment arm?

The original sample size for children enrolled in the intensive biological sampling was based on calculations for an environmental enteric dysfunction substudy. We sampled more children within the treatment arm to maintain sample size and power in the substudy since each environmental enteric dysfunction outcomes were analyzed at each separate time point, and we were sampling from the same study base. This telomere measurement was nested within the environmental enteric dysfunction substudy, and we analyzed any available biological samples in the control and N+WSH arms.

To clarify, we changed the language in Figure 1 – Follow-up to “New children measured”.

8) It's not clear how long the intervention operated, e.g., how long health promoters visited the compounds, whether the structural changes continued to be available throughout. I know that there is a citation to another paper that describes it in more detail, but we need a bit more detail here to understand why this intervention might influence TL.

The interventions operated for 2 years in total.

To address the reviewer’s comment, we have revised the first sentence of the first paragraph in the Results section: “The objective of the WASH Benefits trial was to compare the effects of individual and combined interventions on child health in the first two years of life – the critical window to prevent growth faltering (Arnold et al., 2013).”

Please also refer to the first paragraph of the Results section:“The substudy team visited 996 index children after 1 year of intervention (Y1) and 1021 children after 2 years of intervention (Y2) in the control and N+WSH arms only.”

We have revised this sentence in the Materials and methods section to address the duration of health promoter visits: “Community health promoters visited study compounds in the intervention arms at least once per week during the first 6 months and at least once biweekly to promote behaviors.”

We have revised this sentence in the Results section to address the duration of the observed structural changes: “Measures of intervention adherence (including observed hardware availability) were high (over 80%) and sustained throughout Y1 and Y2 (S. P. Luby et al. in review).”

Methodological concerns relative to measurement of telomere length:

1) More information and data should be included on how the telomere length (in bps) was calculated. 6763 bp is short for infants.

We have added the requested information and data to the Materials and methods section, and discussed the short infant TL. Please see our response to comment #4 under “Essential revisions” for details.

2) Has it been validated to wait so long from sample collection to measurement of TL (subsection “Procedures”, last paragraph)? If not, this is a limitation.

We have not assessed the impact of blood storage time on telomere length measurement, although the majority of the published literature on telomere length in human studies used archived blood (Eisenberg et al., 2017). In the Eisenberg et al. study, the duration of time between venous blood collection and TL measurement was 7 years (Eisenberg et al., 2017). In contrast, in our study, the duration of time from sample collection to TL measurement ranged from 8 to 32 months. We have added it as a limitation to the Discussion section.

The new language reads: “The lithium heparin additive in the whole blood may have inhibited the qPCR reaction and the long duration of time from sample collection to TL measurement may have affected TL, but any systematic measurement errors would likely affect both study groups equally. These factors may provide potential explanations for the shorter average TL observed in this study compared to other studies (Factor-Litvak et al., 2016; Frenck et al., 1998; Wojcicki, Heyman, et al., 2016).”

3) Were there differences in the month in which the treatment and control groups were measured? If so, there are known seasonal differences in TL at leukocytes at least (Rehkopf et al., AJHB 2014). Please discuss.

When we plot TL versus month of data collection for the treatment and control groups (see Author response image 1), we do not see evidence of seasonal changes. In the fully adjusted model, one of the pre-specified covariates was month of data collection to account for seasonal changes.

Author response image 1.

Author response image 1.

Open in a new tab

We have added new language to the Results section: “Although a previous study reported seasonal variation in TL (Rehkopf et al., 2014), we did not observe seasonal changes within this study.”

4) The blood sample for telomere length (by PCR) was stated to be lithium heparin, yet heparin inhibits PCR reactions. Sometimes this can be overcome by use of heparinise. Can the authors please clarify the type of specimen used, and if it was Li-heparin how this may have impacted the key results.

The original text in the Materials and methods section states the type of specimen used: “For each child, trained icddr,b staff collected a 5 ml venipuncture sample from the antecubital area of the arm into a Sarstedt S-monovette lithium heparin collection tube […] Genomic DNA was extracted from heparin-anti-coagulated whole blood stored at -80°C using QIAamp DNA Mini Kit (QIAGEN, Hilden, Germany).”

We did not use heparinase in the PCR reaction. The results that may have been affected by the lithium heparin are the relative telomere lengths for each individual. Since lithium heparin tubes were used for the collection of all blood samples, we would expect any impact of the heparin on the qPCR reaction to affect both the control and intervention groups similarly. Therefore, we would not expect the key differences between the control and intervention groups to be impacted by the heparin.

We added this heparin issue as a potential limitation to the Discussion section. The new sentence reads: “The lithium heparin additive in the whole blood may have inhibited the qPCR reaction and the long duration of time from sample collection to TL measurement may have affected TL, but any systematic measurement errors would likely affect both study groups equally. These factors may provide potential explanations for the shorter average TL observed in this study compared to other studies (Factor-Litvak et al., 2016; Frenck et al., 1998; Wojcicki, Heyman, et al., 2016).”

5) How measuring TL in leucocytes might influence the results, if the proposed mechanism of the intervention is largely infectious or nutritional? This should be listed as a limitation also, since TL measured in different tissues might yield different results based on a given intervention's mechanisms.

Thank you for raising this issue. If the mechanism of the intervention is through the promotion of better immune development, we would expect to observe differences in leukocyte TL (e.g., due to T-cell proliferation), but not in less proliferative tissues such as muscle. We added this issue to the limitations paragraph of the Discussion section.

The new language reads: “We only measured TL in whole blood, which might yield different results compared to TL measured in less proliferative tissue types (e.g., muscle or fat) (Daniali et al., 2013).”

Concerns related to analysis and interpretation:

1) A critical issue is that unadjusted results are presented as the primary, and it is stated that the adjusted and IPCW results are "similar" (Results, fourth paragraph). This is not actually the case. While the point estimates are similar, they are no longer statistically significant, for the most part, in the adjusted and IPCW models. It would be more appropriate to conclude that this study showed less strong effects of the intervention on TL, than the authors imply. It seems inappropriate to consider the unadjusted analysis as primary, since so much differential loss to follow-up occurred.

[repeated from Essential revisions comment #3 response, above]Typically, in a clinical trial, adjusted analyses (including IPCW) are pursued for two main reasons: 1) to remove bias from chance imbalances in group characteristics despite randomization or differential loss to follow-up, and 2) to increase the precision of the effect estimate (Pocock et al., 2002). Balancing these objectives is a classic bias-variance tradeoff in statistics, and our approach, pre-specified in our analysis plan (https://osf.io/cjjwa/), has been to focus first on the issue of bias: namely, do we see evidence for any change in effect after adjustment? Then, if we see no evidence for bias, focus on the unadjusted analysis for our inference. This emphasis follows guidelines from leading statisticians in clinical trials, for example:

“Experiments offer more reliable evidence on causation than observational studies, which is not to gainsay the contribution to knowledge from observation. Experiments should be analyzed as experiments, not as observational studies. A simple comparison of rates might be just the right tool, with little value added by ‘sophisticated’ models.”

Freedman DA. Statistical models for causation – What inferential leverage do they provide? Eval Rev. 2006;30: 691–713.

“Experience shows that for most clinical trials, analyses which adjust for baseline covariates are in close agreement with the simpler unadjusted treatment comparisons. This is because (a) the randomization usually results in well balanced treatment groups, and (b) most covariates are not strongly related to the outcome.”

Pocock SJ, Assmann SE, Enos LE, Kasten LE. Subgroup analysis, covariate adjustment and baseline comparisons in clinical trial reporting: current practice and problems. Stat Med. 2002;21: 2917–2930.

As the reviewer notes, the point estimates across analyses are very similar, which we find reassuring because it is consistent with good balance from randomization and outcome measurements missing at random (with respect to measurable characteristics). As the reviewer also notes, the confidence intervals widen slightly (and thus P-values increase slightly) as we move from the unadjusted analysis to the adjusted analyses. An analysis will only gain precision if all covariates are very strongly associated with the outcome (Pocock et al. 2002). Although our pre-specified algorithm pre-screened covariates that were associated with the outcome (a necessary condition for confounding), the adjusted analyses still lost a small amount of precision – suggesting that our conservative inclusion to reduce all possible bias led to a small loss in precision in this study. Given the similarity of the point estimates across analyses (Table 3), we feel that the unadjusted results, which rely only on randomization, are the most parsimonious summary of the trial.

We summarized our rationale in the Materials and methods section of our original submission: “Randomization led to balance in observed covariates across arms, so, in accordance with our pre-specified analysis plan, we relied on the unadjusted analysis as our primary analysis for the TL outcomes.”Tables 1 and 2 provide evidence that the groups with Y1 and Y2 TL outcomes were balanced across enrollment characteristics and that the group lost to follow up did not differ in enrollment characteristics from those with outcomes.

In the Results section, we addressed the reviewer’s concerns regarding the specific statement about unadjusted, adjusted, and IPCW analyses being “similar”. We clarified this statement for the change in TL from Y1 to Y2 result by emphasizing which model was significant: “Unadjusted, adjusted, and inverse probability of censoring weighting analyses generated similar estimates but only the unadjusted model was significant (Table 3).”Although we stated that the difference between the control and intervention arms in the change in relative TL from Y1 to Y2 was borderline significant (in the Results section), we agree that this borderline significant result is overemphasized in the Discussion section. Therefore, we have softened the language and reframed the discussion of this result in the Discussion section in accordance with guidelines for the interpretation of borderline significant results within clinical trials (Hackshaw and Kirkwood, 2011). Please see our response to comment #4 under “Concerns related to analysis and interpretation” for the revised language in the Discussion section.

2) Table 1 does not present the results for whether baseline covariates were statistically significantly different from one another for the two groups. Were there differences in sociodemographic characteristics between the two groups at Y1 and Y2? I.e., differential loss to follow-up that might explain the differences in TL and attrition? This is presented in a table but should be discussed here.

In a clinical trial, any imbalance in baseline characteristics despite randomization is, by definition, due to chance. The issue was best summarized by Doug Altman:

“In a clinical trial in which treatment allocation was properly randomized, a difference of any sort between the two groups at the time of entry to the trial will necessarily be due to chance, since randomization precludes any external influences (biases) on which subjects receive which treatments […] Performing a significance test to compare baseline variables is to assess the probability of something having occurred by chance when we know that it did occur by chance”.

Altman DG. Comparability of Randomised Groups. Statistician. 1985;34: 125–136.

Instead, Altman recommends displaying enrollment characteristics of the groups in a table to enable the reader to assess balance subjectively. Subjective assessment requires “prior knowledge of the prognostic importance of the variables being compared”. (Altman, 1985)

To address whether there were differences in enrollment characteristics between those with TL outcomes at Y1 versus Y2, we revised Table 1 to include treatment group means by year of measurement.

[repeated from Essential revisions comment #2 response, above] We agree with the reviewers that differential loss to follow-up could potentially be a source of bias in this trial. This is an issue we had anticipated from the trial’s conception and included in our pre-specified and pre-registered analysis plans for the trial (all outcomes, not just telomere outcomes). In our original submission to eLife we included results from an inverse probability of censoring weighted (IPCW) analysis, which re-weights the analysis population to reflect the enrollment population using measured characteristics. The IPCW approach follows best practices for removing potential bias owing to missing outcome measurements in trials, according to a definitive methodologic review conducted by the National Research Council (Little et al., 2012). In this study, the IPCW analysis led to very similar inference as the adjusted and unadjusted analyses (original submission Table 4, revised submission Table 3), which suggests that outcomes were missing at random with respect to measurable characteristics and the missing outcomes are not a likely source of bias in the study.

In the original submission, we included Table 2 featuring the enrollment characteristics of households with TL measurement at Year 1 versus those lost to follow-up at Year 2. We have added differential loss to follow-up as a potential limitation in the Discussion section.

The new language in the Discussion section reads: “Differential loss to follow-up in the control and intervention arms could potentially introduce bias if differential loss to follow-up was caused by unmeasured characteristics. However, our IPCW analysis showed that bias from loss to follow-up was unlikely based on a large set of measurable characteristics.”

3) Many more children withdrew from the control group. This again highlights the need to present study characteristics to examine possible differential loss to follow-up. And this should be included as a major limitation, in that unmeasured characteristics may be driving these differences and the subsequent results. It also suggests that perhaps the adjusted analyses should be the primary ones.

Please see our response to comment #2 “Concerns related to analysis and interpretation” above.

Thank you for raising this issue. To address the differential loss to follow-up comment, we added this language to the Discussion section: “Differential loss to follow-up in the control and intervention arms could potentially introduce bias if differential loss to follow-up was caused by unmeasured characteristics. However, our IPCW analysis showed that bias from loss to follow-up was unlikely based on a large set of measurable characteristics.”

4) Again, it would be more appropriate to frame this as a study that found null effects except for Year 1 differences, and to discuss why there may have been no effect observed. Moreover, there is not enough discussion of the hypothesized pathways linking adversity and TL, nor on how this intervention addresses adversity as opposed to nutrition and sanitation, which are very different exposures.

[repeated from Essential revisions comment #3 response, above] In the Discussion section, we have reframed this study as a study that found significant results at Y1, borderline significant/null effects in the difference in the relative change in TL between Y1 and Y2 among the two groups, and null effects at Y2.

To emphasize this point, we have added the borderline significant effects in the difference in the relative change in TL between Y1 and Y2 among the two groups and the null effects at Y2 to the first paragraph of the Discussion section: “The difference between the intervention and control arms in the change in relative TL from Y1 to Y2 was borderline significant, and at Y2, we found no difference in relative TL between the arms.”

In accordance with the guidelines for the interpretation of borderline significant results in clinical trials (Hackshaw and Kirkwood, 2011), we have softened the language of our interpretation and put forth alternative explanations.

The revised language in the Discussion emphasizes the borderline aspect of the change in TL result and enables the reader to interpret whether or not the difference is biologically meaningful and/or if the mechanisms proposed are plausible: “After the initial period of rapid linear growth during Y1, our borderline significant finding that telomere length increased in the intervention arm and decreased in the control arm from Y1 to Y2 seems to represent a preview of the TL trajectories that are set: we hypothesize that the intervention children will experience slower lifetime TL attrition and more physiological resistance to stress-related diseases (Juster et al., 2010). […] The smaller than expected treatment effect and the borderline p-value indicate that the trial was slightly underpowered to detect differences in the change in relative TL from Y1 to Y2 among the intervention and control groups.”

Additionally, we added a new paragraph with an alternative interpretation if the reader considers the borderline result to be null: “Alternatively, our finding of a modest, borderline impact from Y1 to Y2 may suggest that the effect of the interventions is diminished in children over 1 year of age. […] This interpretation would also be consistent with no differences in relative TL observed at Y2 between the two groups, underscoring the importance of targeting interventions early in life during the sensitive period when they are likely to have the largest impact on childhood TL.”

We have amended the Discussion section to remove the focus on adversity (as suggested by the reviewer in comment #7 under “Concerns related to analysis and interpretation” below). Furthermore, we have expanded on our original discussion of mechanisms by including a new paragraph in the Discussion on the hypothesized pathways linking nutrition and WSH exposures and childhood TL and growth. Please see our response to comment #5 under “Concerns related to analysis and interpretation” (below) for more details.

5) The authors should postulate mechanisms by which telomere shortening improves growth.

In the original Discussion section, we hypothesized that telomere shortening may reflect improved growth (e.g., a biomarker of growth) and did not imply that telomere shortening would necessarily be considered a cause of improved growth (if anything, our hypothesis based on the results would imply causality in the other direction: improved growth → shorter telomeres). We have included language in the Discussion to further clarify our hypothesis (please see new text below). Our study was not designed to examine the potential causal pathway between telomere shortening and improved growth. Therefore, we prefer not to postulate mechanisms for a relationship that we have not yet examined or proposed.

Instead, to address this point and comment #5 under “Essential revisions” we have expanded the discussion on potential mechanisms for improved TL and growth observed in the intervention arm by adding this paragraph to the Discussion section: [repeated from Essential revisions comment #5 response, above] “The accelerated TL attrition during the first year of life in the intervention arm may also reflect better linear growth. […] Forthcoming WASH Benefits studies focused on child inflammation and enteric pathogen burden will provide evidence on mechanistic pathways leading to different TL, growth, and development outcomes.”

6) If telomere length shortening is associated with better immune functioning and the infants had improved hygienic conditions, then it is counterintuitive when considered in context with germ theory of disease. This should be amended and discussed.

We have not made this amendment because our interpretation is that the observed results and hypothesized relationship between hygienic conditions, immune function, and telomere length shortening are all internally consistent with the germ theory of disease. The germ theory of disease states that infectious diseases are caused by the presence of pathogenic microorganisms within the body. As discussed in the Discussion section, the children in the N+WSH intervention arm potentially experienced improved hygienic conditions, which likely limited exposure to microorganisms, as exhibited by the reductions in caregiver-reported diarrhea, soil-transmitted helminth infections, and Giardia duodenalis infections (S. P. Luby et al., A. Ercumen et al., and A. Lin et al. in review). The children in the N+WSH intervention arm likely experienced fewer infections and perhaps had better immune system development, which in turn was reflected by shorter telomeres.

7) The authors discuss adversity, which conjures to mind poverty, psychosocial stress, etc. It is inappropriate to talk about adversity because it was not measured via socioeconomic or psychosocial risk factors, so much as nutrition and sanitation. Please clarify what "adversity" means in these prior studies, and how/whether that relates to this particular intervention.

We significantly reframed the Introduction section by removing the focus on adversity and instead, adding two new paragraphs on the potential mechanisms by which nutrition and WSH could affect TL. For more details, please see our response to comment #1 under “Essential revisions” (above).

We have also amended the Discussion to clarify the term “adversity” in prior studies: “This result is contrary to many studies that reported increased TL attrition associated with prenatal psychosocial stress (Entringer et al., 2013; Marchetto et al., 2016), childhood institutional care (Drury et al., 2012), disease (Fitzpatrick et al., 2007; Salpea et al., 2010), and mortality (Cawthon, Smith, O'Brien, Sivatchenko, and Kerber, 2003); however, these prior studies did not examine participants in the same age range (1-2 years) as this study. In high- and low-resource settings, there is a dearth of evidence on the effect of environmental exposures on TL in this age range and its implications for adult health outcomes.”

8) The limitations section needs to be expanded, e.g., lack of process evaluation measures that might clarify mechanistic pathways linking intervention to disease; limited understanding of how different types of exposures might influence TL; differential loss-to-follow-up on unmeasured characteristics that can't be accounted for by IPCW; capturing TL only in a single tissue. What are the implications of the lack of geographic matching for the substudy? What are the implications for generalizability and bias of the exclusion criteria, which eliminated potentially more disadvantaged individuals and therefore may have selected on better outcomes?

Thank you for identifying these additional limitations. We have expanded the limitations section to discuss all of the reviewer’s suggestions except the following points: “lack of process evaluation measures that might clarify mechanistic pathways linking intervention to disease”, “limited understanding of how different types of exposures might influence TL”, and “bias of the exclusion criteria, which eliminated potentially more disadvantaged individuals and therefore may have selected on better outcomes”. The process evaluation measures are described in detail in the main trial publication (S. P. Luby et al. in review), and all analyses were intention-to-treat. The limited understanding of how different types of exposures might influence TL is not a limitation specific to this study, rather it is a limitation of the general TL literature. Therefore, we have included this point in the final two sentences of the Discussion section, but have not highlighted it in the limitations paragraph. The venipuncture exclusion criteria only excluded two children: one child in the control arm at Year 1 and one child in the intervention arm at Year 2. We added this information to the Materials and methods section. Since the number of children excluded is small, we have decided not to include the exclusion criteria as a limitation.

[repeated from Essential Revisions comment #5 response, above] The revised limitations paragraph reads: “This study had several limitations. […] Finally, we did not measure TL at birth (Wojcicki, Olveda, et al., 2016), but we would expect it to be similar for both groups because household characteristics were balanced by randomization.”

9) Only including "significantly related covariates" is not an appropriate way to select covariates. You should include all of those that are theoretically potential confounders. I.e., covariates should be selected conceptually and not empirically. It doesn't seem necessary to include (1) just age child age and sex; just do one fully adjusted model with all relevant covariates (subsection “Statistical parameters”, second paragraph).

Please see our response to comment #1 under “Concerns related to analysis and interpretation” for our broader perspective on adjusted analyses in the context of clinical trials. In regard to this specific point, by definition a covariate can confound a relationship between an exposure (here: randomized treatment) and the outcome if it is associated with the outcome (VanderWeele and Shpitser, 2011). However, it is widely known that covariates can hurt precision in an analysis if they are not strongly associated with the outcome. Therefore, this pre-screening step is a common approach for covariate selection in trials – and epidemiologic studies more broadly – to help avoid losing too much precision in the adjusted analysis through inclusion of covariates that are unassociated with the outcome (and thus could not be confounders). As long as the approach is algorithmic and pre-specified, it is not susceptible to “cherry picking”. Below is an excerpt of an influential article on the analysis of clinical trials that contributed to the development of our pre-specified approach to covariate selection in this trial:

“Some have argued that prespecification of all covariates for adjustment in a single model is the best solution, leading to just one predefined covariate-adjusted analysis [10]. […] While this means the chosen covariates themselves could not be prespecified, a precise predefined statistical strategy for variable selection should overcome somewhat any suspicions that post hoc selection of covariates might be based on subjective criteria.”

Pocock SJ, Assmann SE, Enos LE, Kasten LE. Subgroup analysis, covariate adjustment and baseline comparisons in clinical trial reporting: current practice and problems. Stat Med. 2002;21: 2917–2930.

10) The authors state that they conducted inverse probability weighting, but do not give these details (subsection “Missing outcomes”).

We have added more details on IPCW analysis to the Materials and methods section.

The newly added language reads: “The inverse probability of censoring weighted (IPCW) approach follows best practices for removing potential bias owing to missing outcome measurements in trials, according to a definitive methodologic review conducted by the National Research Council (Little et al., 2012). […] TL differences between the control and intervention group were then estimated using adjusted linear regression weighted by the inverse probability of TL measurement.”

11) For both Tables 1 and 2, it's confusing to have mean (SD) and N (%) in the same column, and the tables overall are clutters with all the "%" signs. Perhaps have mean (SD) for continuous outcomes and just the% for binary outcomes?

As suggested, we have displayed the mean (SD) for continuous outcomes and only the% for binary outcomes. We have retained the% signs for clarity.

12) Table 1: Is this table for the children who had telomere outcomes measured? Please specify in the footnote.

Yes, this table is for children who had telomere outcomes measured.

We have added this language to the footnote: “Enrollment characteristics of households with children who had telomere measurements.”

13) Table 1: At the top left, the table says "No. of compounds." Are the "N =" at the top of the second and third columns therefore number of compounds or number of children? If number of compounds, please present number of children also/instead. Similarly for the top of Table 2.

Yes, these numbers are the number of children.

In Table 1, we have added changed the heading to: “No. of children”

14) Tables 3 and 4 are very redundant. Can't you just add the IPCW column from Table 4 to Table 2? Perhaps put the CI below the point estimate so it can all fit?

We agree and have combined the original Tables 3 and 4 into one table by adding the IPCW column from Table 4 to Table 3.

15) The footnotes of Tables 3 and 4 say that covariates were pre-specified, but the text says that adjustment included significantly related covariates at p < 0.2. Please clarify.

We have clarified the footnote:“Adjusted for pre-specified covariates associated with the outcome (likelihood ratio test P-value <0.2).”

16) Table 5 would look better if girls and boys were listed side by side instead of all in the same column. And actually, presenting it stratified in this way doesn't give us the information we want, which is the point estimate and CI for the interaction term of gender x treatment group. Are the headings for the mean and N columns reversed?

We have made the changes requested and included the point estimate and CI for the interaction term. Please note that the original Table 5 has been renumbered as Table 4.

17) It is not intuitive why the argument for TL measurement is strengthened as a "potential target for future intervention development and evaluation." To the contrary, this suggests that the simple belief that longer TL = healthier needs to be further nuanced, and that additional studies like this one need to be done. We need a better understanding of the human physiology linking different types of socioeconomic and biomedical exposures with different types of telomere-related outcomes, since the existing literature suffers from too much confounding to understand what is actually going on. I'd say the main takeaway from this study is that we have a very limited understanding of how TL in humans actually responds to different types of interventions, and limited understanding of how TL actually influences different aspects of human health.

Thank you for synthesizing the salient points of our study. We strongly agree with the reviewer’s comments. Additionally, there are gaps in the literature on child TL dynamics in health and disease during the first two years of life. As suggested by the reviewer, we have revised and strengthened the concluding sentences of the Discussion based on these points.

The revised language reads: “Although TL was a sensitive outcome that responded to an intervention, this trial underscores our limited understanding of environmental exposures contributing to TL dynamics during early life and the critical physiological pathways linking TL and lifelong health trajectories. Since potential confounding could plague observational studies, evaluating the relationship between modifiable exposures and TL within the context of randomized controlled trials could provide valuable contributions to the field of telomere biology.”

18) The presentation of the results in the Abstract is confusing, and doesn't parallel the presentation in the main manuscript. It would be helpful if the authors stated (1) the difference in TL between treatment and control at year 1; (2) the difference in TL between treatment and control at year 2; and (3) the difference in telomere attrition between treatment and control at year 1 and at year 2. These are the primary outcomes, as I understand them.

We have restructured the Abstract to parallel the results in the main manuscript by adding point 2: difference in treatment and control at Year 2. The original Abstract already included points 1 and 3.

The revised paragraph reads: “Children in the intervention arm had significantly shorter relative TL compared with controls after 1 year of intervention (difference -163 bp, P=0.001). Between years 1 and 2, TL increased in the intervention arm (+76 bp) and decreased in the controls (-23 bp) (P=0.050). After two years, there was no difference between the arms (P=0.305).”

19) Unclear wording: does linear growth refer to telomeres or the child's height/length? (Abstract) If the latter, then this was not examined in this manuscript and should not go in the Abstract.

Linear growth refers to the child’s length/height, the primary outcomes of the main trial. We have deleted this sentence from the Abstract.

Associated Data

    This section collects any data citations, data availability statements, or supplementary materials included in this article.

    Data Citations

    1. Lin A, Mertens A. 2017. WASH Benefits Bangladesh Analysis of Telomere Outcomes. Publicly available at the Open Science Framework (project no. evc98) [DOI]

    Supplementary Materials

    Figure 1—source data 1. Source data and code for Figure 1.
    elife-29365-fig1-data1.zip (109.6KB, zip)
    Table 1—source data 1. Source data and code for Table 1.
    elife-29365-table1-data1.zip (88.4KB, zip)
    Table 2—source data 1. Source data and code for Table 2.
    elife-29365-table2-data1.zip (90.3KB, zip)
    Figure 2—source data 1. Source data and code for Figure 2.
    elife-29365-fig2-data1.zip (85.5KB, zip)
    Table 3—source data 1. Source data and code for Table 3.
    elife-29365-table3-data1.zip (283.9KB, zip)
    Table 4—source data 1. Source data and code for Table 4.
    elife-29365-table4-data1.zip (90.7KB, zip)
    Reporting standard 1
    elife-29365-repstand1.docx (132.3KB, docx)
    Transparent reporting form
    elife-29365-transrepform.pdf (313.1KB, pdf)

    Data Availability Statement

    The WASH Benefits data and code that support the findings of this study are available in Open Science Framework (https://osf.io/evc98/).

    The following dataset was generated:

    Lin A, Mertens A. 2017. WASH Benefits Bangladesh Analysis of Telomere Outcomes. Publicly available at the Open Science Framework (project no. evc98)

    Articles from eLife are provided here courtesy of eLife Sciences Publications, Ltd

    RESOURCES