Evidence for Genetic Susceptibility to Developing Early Childhood Diarrhea among Shantytown Children Living in Northeastern Brazil

Early childhood diarrhea (ECD) is a leading cause of child mortality and morbidity globally and causes more than 1.34 million deaths per year.1,2 Children who survive repeated episodes of diarrhea often have sustained growth shortfalls and impairments in physical fitness, cognition and school performance.3–6 Early childhood diarrhea is associated with a wide range of microbes and pathogenesis.7,8 However, the human genes responsible for variation in susceptibility to acquiring pathogens associated with diarrhea remain poorly explored.

Host genetics have been increasingly implicated in susceptibility to specific parasites. These parasites include Ascaris lumbricoides, Cryptosporidium spp., and Entamoeba histolytica.9–14 However, little is known about the importance of host genetics for risk of developing ECD. Knowledge of the genetic determinants of host susceptibility to ECD may facilitate development of novel approaches for control of the disease.

This study applied a variance component approach to quantify the relative contribution of genetic and environmental components to risk for ECD. We used extended pedigree information, global positioning mapping (GPS) of household residence and ECD data generated for children living in two favelas in Fortaleza, Brazil. The study protocol and procedures for obtaining informed consent were approved by the University of Virginia Institutional Review Board for Health Sciences Research, the local Institutional Review Board of the University of Ceara in Fortaleza, and the Brazilian Ministry of Health.

Family information was gathered from two urban shantytowns (Parque Universitario and Gonçalves Dias) located within the capital city of Fortaleza, Ceara, in northeastern Brazil. Pregnant women in these communities were identified, and enrollment of newborns was initiated in 1989. Each child in the study was visited three times a week by a study nurse who recorded diarrheal illness, sociodemographics data, and breastfeeding information. Children who had diarrhea were visited daily until 48 hours after resolution of illness. Mothers were asked to provide detailed clinical information about the illness including stool characteristics and other symptoms such as fever, vomiting, and dehydration.

Extended pedigrees were reconstructed on the basis of information gathered for all household participants in the two favelas described according to the approach of Williams-Blangero and Blangero.15

The World Health Organization criteria to determine diarrhea episodes were used for our analyses. Diarrhea was defined as ≥ 3 liquid stools in the preceding 24-hour period. An episode of diarrhea was defined as lasting ≥ 1 day and separated from another episode by ≥ 2 days without diarrhea. Household locations were mapped by using GPS. Distance information was included in the variance component models to assess risk for ECD caused by geography and common living environment.

We used the Sequential Oligogenic Linkage Analysis Routines (SOLAR) program to assess the amount of variance accounted for by genetic relatedness in developing ECD. This program calculates variance in risk for ECD attributable to genetic factors. It also uses a variance component approach to partition the total phenotypic covariance (Ω) of the traits into additive genetic (σ²_G) and environmental (σ²_E) variance components according to the formula Ω = 2Φσ²_G + Iσ²_E, where Ω is a matrix of phenotypic covariance between pairs of persons, 2Φ is a matrix of expected allele sharing on the basis of degree of relationship, and I is an identity matrix. The environmental component includes factors that are not explicitly measured such as socioeconomic status, measurement errors, and dominance or other non-additive genetic factors. The additive heritability (h²) of a trait represents the portion of the total phenotypic variance σ²_P accounted for by the additive genetic variance formula h² = σ²_G/σ²_P.

To determine the significance of heritability, a null model in which the additive genetic variance (σ²_G) for the trait equals zero is compared with another model, where σ²_G is estimated by maximum likelihood methods. Because σ²_G is tested on its boundary, the likelihood ratio test statistic is distributed as a ½:½ mixture of a chi-square distribution with one degree of freedom and a point mass at zero.

The decomposition of the phenotypic covariance is readily extended to include other random effects. We tested models that enabled two such effects: the effect of common household (σ²_C), and the effect of physical proximity (σ²_PX) based on GPS data. The former variance component was structured by a matrix whose elements were 1 if the children shared a household or 0 otherwise. The latter variance component was structured by a proximity matrix computed as follows: for each pair of children i,j with ECD phenotypes, a proximity score was calculated as px_ij = exp(–d_ij), where d_ij is the physical distance in hectometers between the children's households. Thus, the proximity score ranged from 1 for children sharing a household to effectively zero at a distance ≥ 1 km.

There were 277 children for whom ECD and GPS data were available. Children characterized for ECD belonged to 127 pedigrees, which ranged from 3 to 188 family members and had a mean (SEM) of 25.72 (2.3) members (Table 1). Pedigrees contained a mean (SEM) of 10.13 (0.87) nuclear families, which ranged from 1 to 72 nuclear families per pedigree. Each of these contained approximately equal numbers of full (mean = 2.44, SEM = 0.11), maternal (mean = 2.83, SEM = 0.11), and paternal (mean = 2.51, SEM = 0.11) sibships. Most (72.9%) pedigrees included at least one full sibship.

The degree of genetic relatedness among children in the study is delineated in Table 2. There was a high proportion (36.5%, 118 of 323 pairwise relationships) of first cousin pairs within this population. Because cousins typically lived in separate households, these and cousins of other degrees are a particularly informative group for distinguishing between genetic and environmental influences.

There was a high prevalence (89.5%, 248 of 277) of diarrhea before two years of age. More than half (50.5%, 140 of 277) the sampled children experienced ≥ 4 episodes, and 27.4% (76 of 277) had ≥ 7 or more episodes. Most mothers (82.5%, 212 of 257) had not completed elementary school. The average child was breastfed until 5.4 months of age (range = 0–24 months). Household crowding averaged 5.6 family members sleeping in a single home and ranged from 2 to 16. Average per capita income per month was approximately US $113 and ranged from none to US $755.

The eight models (A–H) used to assess the genetic versus environmental contribution to ECD are described in Table 3. A model was also run to assess the likelihood that the phenotypic variance attributed to genetics shows a more dominant or additive inheritance pattern. Variance component estimates, standard error of variance estimate, and log_e likelihood for each model are reported. The test column in Table 3 indicates which models where compared, and the P values refer to the probability that the models compared are not significantly different from one another; P < 0.05 was considered significant.

The polygenic model (model B), which includes only additive genetic (h²) and environmental (e²) random effects, showed that relatedness accounts for 0.86 (SEM = 0.18) of ECD variance. This finding was highly significantly different (P = 0.000001) from the sporadic (random environmental) model (model A). Model C shows some evidence of a dominance effect, which interestingly absorbs all the random environmental effect (e²). However, Model C was not significantly different from Model B (P = 0.217637).

Model D indicates that living in a common household has a significant effect on ECD (P = 0.000031), but once polygenic factors were included in Model E, the effect of household became null. The P value for Model E (P = 0.500000) indicates a non-significant difference between the combined effect of Household/Polygenic compared with Polygenic alone. Model F shows household proximity alone has a large effect size (0.74 ± 0.08), which was reduced to 0.25 ± 0.13 after polygenic factors were accounted for, as shown in Model G. The P value for Model G (P = 0.022217) indicates that proximity accounts for a significant proportion of variation in ECD, independent of polygenic factors.

The variance component analysis of ECD (Table 3) shows that a statistically significant and substantial proportion of the variation in number of episodes of diarrhea experienced in the first two years of age in these favela populations is attributable to genetic factors. Overall, genetic relatedness accounted for 0.54 ± 0.19 of the phenotypic variance of ECD, and geographic relatedness (proximity of residence) accounted for an additional 0.21 ± 0.09 (P < 0.0001). Furthermore, educational status of mothers, breastfeeding practices, and household crowding did not significantly account for variance in ECD once genetic relatedness and geographic proximity of residence were accounted for. Data not shown.

The study children are known to experience substantial exposure to contaminated drinking water, live in unsanitary conditions, and experience a high burden of infection.18,16 Contaminated water sources and inadequate water storage are highly prevalent in diarrhea-endemic areas.17 This finding, in combination with household crowding, can facilitate spread of enteric pathogens.18,19 In our cohorts, enteroaggregative Escherichia coli, Cryptosporidium, and Giardia spp. are leading causes of recurrent diarrhea in children in their first two years of life and are easily spread from person-to-person.20,21

Susceptibility to infection might be associated with disruption of intestinal function by the pathogen, leading to inflammation or tissue damage, which is also likely to be genetically influenced. Jiang and others studying single-nucleotide polymorphisms in the interleukin-8 promoter region in American travelers to Mexico found that persons with symptomatic enteroaggrgative E. coli had higher frequencies of AA genotype at the –251 position and increased levels of fecal interleukin-8 associated with intestinal epithelial disruption and fluid secretion.21

Our finding that genetic factors account for an estimated 54% of the variability in ECD suggests that further genetic studies are warranted. Several underlying phenotypes may be incorporated in the ECD phenotype subjected to analysis. For example, the variation in ECD observed in this study captures variation in acquiring infection and subsequent inflammatory or secretory responses leading to overt diarrhea. Future studies should include underlying phenotypic pathologies, such as intestinal inflammation, specific etiologies, and other phenotypic characteristics such as illness durations, related to ECD. Because co-morbidities and cumulative enteric diseases are common in children from disease-endemic areas in the developing world, a composite of genetic components of susceptibility to diverse pathogens warrant future studies of specific etiologies and pathotypes of enteric infection.

Our findings support the importance of taking the next step in investigating this topic by investing in increasingly economically feasible genome wide association or case–control studies designed to identify functional genetic variants associated with ECD in northeastern Brazil. Identification of specific genetic contributions could be used to illuminate the underlying etiologies and pathogenesis of ECD and may suggest new mechanisms to target with novel therapies.