Module 1: The Exposome — Motivation, Biology, and NHANES

Conducting Exposome-Wide Association Studies

Chirag J Patel

Welcome to the Course

Conducting Exposome-Wide Association Studies (ExWAS)

  • 7-module online course
  • Learn the statistics and biology behind ExWAS
  • Hands-on R programming with real NHANES data
  • Culminates in running and interpreting your own ExWAS

Course Overview

Module Topic
1 The Exposome: Motivation, Biology, and NHANES
2 R and Tidyverse Foundations
3 Statistical Foundations for ExWAS
4 The nhanespewas Package
5 Conducting an ExWAS
6 Interpreting ExWAS Results
7 Advanced Topics

What is the Exposome?

“The exposome encompasses the totality of human environmental exposures from conception onwards, complementing the genome.”

— Christopher Wild, 2005

  • The genome is the complete set of DNA
  • The exposome is the complete set of environmental exposures
  • Together, they shape health and disease

The E-G-P-D Framework

Our conceptual model links:

  • Environment (exposures): chemicals, nutrients, pollutants, lifestyle
  • Genome: inherited genetic variation
  • Phenotype: measurable traits (BMI, glucose, blood pressure)
  • Disease: clinical outcomes (T2D, cardiovascular disease)

\[E + G \rightarrow P \rightarrow D\]

Understanding the exposome means understanding E in this framework.

Why Study the Exposome?

The rise of chronic disease:

  • Type 2 diabetes prevalence has increased dramatically over the past 40 years
  • Obesity rates have tripled since the 1970s
  • Genetics alone cannot explain these rapid changes
  • Environmental factors — diet, chemicals, lifestyle — are key drivers

CDC Diabetes Atlas: https://www.cdc.gov/diabetes/php/data-research/index.html

Genetic vs. Environmental Contributions

  • Genome-Wide Association Studies (GWAS) have identified hundreds of loci for complex traits
  • Yet GWAS variants typically explain <20% of trait heritability
  • The “missing heritability” problem suggests environmental factors play a large role
  • ExWAS is the environmental analogue of GWAS

What is an ExWAS?

Exposome-Wide Association Study (ExWAS):

  • Systematically test associations between many environmental exposures and a health outcome
  • Analogous to GWAS: instead of scanning SNPs, scan exposures
  • Data-driven, hypothesis-generating approach

First described in: Patel et al. 2010, PLoS ONE

ExWAS: The Concept

  1. Measure a large panel of environmental exposures (E₁, E₂, …, Eₘ)
  2. Measure a phenotype or disease outcome (P or D)
  3. Test each exposure-phenotype association individually
  4. Correct for multiple testing
  5. Replicate findings in independent data

The National Health and Nutrition Examination Survey (NHANES)

  • Conducted by the US CDC / National Center for Health Statistics (NCHS)
  • Representative sample of the non-institutionalized US population
  • Continuous since 1999, released in 2-year cycles
  • Combines interviews, physical examinations, and laboratory tests

https://www.cdc.gov/nchs/nhanes/index.htm

Why NHANES for ExWAS?

  • Hundreds of environmental biomarkers measured in blood and urine
  • Linked to health phenotypes (BMI, glucose, lipids, blood pressure)
  • Demographic and socioeconomic data available
  • Survey weights enable nationally representative estimates
  • Publicly available and well-documented

NHANES Sampling Design

NHANES uses a complex, multi-stage probability sampling design:

  1. Primary Sampling Units (PSUs): counties or groups of counties
  2. Strata: geographic/demographic divisions within PSUs
  3. Households: selected within strata
  4. Individuals: selected within households

This design requires special statistical methods (Module 3).

NHANES Survey Cycles

Cycle Years Key Variables
1 1999-2000 Core demographics, lab biomarkers
2 2001-2002 Expanded environmental panel
3 2003-2004 Continued monitoring
4 2005-2006 Additional pollutants
… … Ongoing

Our training data uses cycles 1-2; testing data uses cycles 3-4.

Types of Exposures in NHANES

Biomarkers of environmental exposure:

  • Metals: lead (Pb), cadmium (Cd), mercury (Hg)
  • Persistent organic pollutants: PCBs, dioxins, pesticides
  • Volatile organic compounds: benzene metabolites
  • Nutrients: vitamins, minerals, fatty acids
  • Lifestyle: cotinine (smoking), caffeine

Measured in blood and urine samples.

Let’s Load Our Data

load('../data/nh.Rdata')

This .Rdata file contains:

  • NHData.train: training dataset (NHANES 1999-2002)
  • NHData.test: testing dataset (NHANES 2003-2006)
  • ExposureList: array of exposure variable names
  • ExposureDescription: dictionary of exposure variables
  • demographicVariables: dictionary of demographic variables

Exploring the Training Data

dim(NHData.train)
[1] 21004   193
ncol(NHData.train)
[1] 193

We have 21004 individuals and 193 variables in the training set.

How Many Exposures?

length(ExposureList)
[1] 160

We have 160 environmental exposures to test.

Exposure Descriptions

head(ExposureDescription, 10) %>%
  kable() %>%
  kable_styling(font_size = 9)
tab_name tab_desc series module var var_desc tab_desc_ewas analyzable var_desc_ewas binary_ref_group var_desc_ewas_sub comment_var version_date is_comment is_weight is_questionnaire is_ecological is_binary is_ordinal categorical_ref_group categorical_levels
52 l06bmt_c Blood Lead, Cadmium, Mercury 2003-2004 laboratory LBXBCD Cadmium (ug/L) heavy metals 1 heavy metals NA NA NA 2009-11-13 0 0 0 0 0 NA NA NA
54 l06bmt_c Blood Lead, Cadmium, Mercury 2003-2004 laboratory LBXBPB Lead (ug/dL) heavy metals 1 heavy metals NA NA NA 2009-11-13 0 0 0 0 0 NA NA NA
56 l06bmt_c Blood Lead, Cadmium, Mercury 2003-2004 laboratory LBXTHG Mercury, total (ug/L) heavy metals 1 heavy metals NA NA NA 2009-11-13 0 0 0 0 0 NA NA NA
58 l06bmt_c Blood Lead, Cadmium, Mercury 2003-2004 laboratory LBXIHG Mercury, inorganic (ug/L) heavy metals 1 heavy metals NA NA NA 2009-11-13 0 0 0 0 0 NA NA NA
60 l06cot_c Cotinine 2003-2004 laboratory LBXCOT Cotinine (ng/mL) biochemistry 1 cotinine NA NA NA 2009-11-13 0 0 0 0 0 NA NA NA
61 l06hm_c Heavy Metals 2003-2004 laboratory URXUBA Barium, urine (ng/mL) heavy metals 1 heavy metals NA NA URDUBALC 2009-11-13 0 0 0 0 0 NA NA NA
62 l06hm_c Heavy Metals 2003-2004 laboratory URXUBE Beryllium, urine (ng/mL) heavy metals 1 heavy metals NA NA URDUBELC 2009-11-13 0 0 0 0 0 NA NA NA
63 l06hm_c Heavy Metals 2003-2004 laboratory URXUCD Cadmium, urine (ng/mL) heavy metals 1 heavy metals NA NA URDUCDLC 2009-11-13 0 0 0 0 0 NA NA NA
64 l06hm_c Heavy Metals 2003-2004 laboratory URXUCO Cobalt, urine (ng/mL) heavy metals 1 heavy metals NA NA URDUCOLC 2009-11-13 0 0 0 0 0 NA NA NA
65 l06hm_c Heavy Metals 2003-2004 laboratory URXUCS Cesium, urine (ng/mL) heavy metals 1 heavy metals NA NA URDUCSLC 2009-11-13 0 0 0 0 0 NA NA NA

Key Phenotypes

Our primary phenotypes related to Type 2 Diabetes:

Variable Description Units
BMXBMI Body Mass Index kg/m²
LBXGLU Fasting Blood Glucose mg/dL 
NHData.train %>%
  summarize(
    mean_bmi = mean(BMXBMI, na.rm = TRUE),
    sd_bmi = sd(BMXBMI, na.rm = TRUE),
    mean_glucose = mean(LBXGLU, na.rm = TRUE),
    sd_glucose = sd(LBXGLU, na.rm = TRUE)
  ) %>% kable(digits = 2) %>% kable_styling()
mean_bmi sd_bmi mean_glucose sd_glucose
24.79 7.03 100.51 32.32

Demographic Variables

head(demographicVariables, 10) %>%
  kable() %>%
  kable_styling(font_size = 9)
var description
SEQN ID
RIDAGEYR age (years)
female female sex
male male sex
black Black
mexican Mexican American
other_hispanic Other Hispanic
other_eth Other Race/Ethnicity
SES_LEVEL SES Tertile (0, 1, 2)
education Education (< HS, =HS, >HS)

Key Demographic Variables

Variable Description
RIDAGEYR Age in years
male / female Sex (binary indicators)
white, black, mexican, other_hispanic, other_eth Race/Ethnicity (binary indicators)
INDFMPIR Income-to-poverty ratio
SES_LEVEL SES tertile (0, 1, 2)
SDMVPSU Primary sampling unit
SDMVSTRA Stratum
WTMEC2YR 2-year survey weight

Variable Types in Our Data

NHData.train %>%
  select(BMXBMI, LBXGLU, RIDAGEYR, male, black, INDFMPIR) %>%
  summarize(across(everything(), class)) %>%
  pivot_longer(everything(), names_to = "variable", values_to = "type") %>%
  kable() %>% kable_styling()
variable type
BMXBMI numeric
LBXGLU numeric
RIDAGEYR numeric
male numeric
black numeric
INDFMPIR numeric

Exploring an Exposure: Blood Lead

ggplot(NHData.train, aes(x = LBXBPB)) +
  geom_histogram(bins = 50, fill = "steelblue", color = "white") +
  labs(x = "Blood Lead (ug/dL)", y = "Count",
       title = "Distribution of Blood Lead Levels") +
  theme_minimal()

Challenges of Exposome Research

  1. Multiple testing: hundreds of tests inflate false positives
  2. Correlation among exposures: exposures are not independent
  3. Confounding: demographic and socioeconomic factors affect both exposure and outcome
  4. Measurement error: biomarkers are imperfect proxies of true exposure
  5. Reverse causation: disease may affect exposure levels
  6. Temporal variability: single time-point measurement

The Multiple Testing Problem

  • Testing 160 exposures simultaneously
  • At \(\alpha = 0.05\), we expect ~8 false positives by chance
  • Solutions:
    • Bonferroni correction: \(\alpha / m\) (conservative)
    • Benjamini-Hochberg FDR: controls false discovery rate (less conservative)
  • More detail in Module 3

Correlation Among Exposures

  • Environmental exposures are often correlated
    • Smokers have higher cadmium AND lead
    • Dietary nutrients co-occur in food patterns
    • Pollutants share environmental sources
  • Correlated exposures complicate interpretation
  • Cannot isolate the “causal” exposure from correlated panel

Confounding

  • A confounder is a variable that influences both the exposure and the outcome
  • Example: age affects both lead levels (older cohorts had more lead exposure) and BMI
  • Must adjust for confounders in regression models
  • Common confounders: age, sex, race/ethnicity, socioeconomic status

Key References

  1. Wild CP. Complementing the genome with an “exposome.” Cancer Epidemiol Biomarkers Prev. 2005.
  2. Patel CJ, Bhattacharya J, Butte AJ. An Environment-Wide Association Study (EWAS) on Type 2 Diabetes Mellitus. PLoS ONE. 2010.
  3. Patel CJ, et al. A database of human exposomes and phenomes from the US NHANES. Scientific Data. 2016.
  4. Chung MK, et al. The exposome and exposome-wide association studies. Exposome. 2024.

What’s Next?

Module 2: R and Tidyverse Foundations

  • The pipe operator and dplyr verbs
  • Data visualization with ggplot2
  • Linear regression with broom
  • Building pipelines for many regressions

Summary

  • The exposome represents the totality of environmental exposures
  • ExWAS systematically scans exposures for associations with health outcomes
  • NHANES provides a rich, publicly available dataset for ExWAS
  • Key challenges: multiple testing, confounding, correlation, measurement error
  • This course will teach you to conduct, interpret, and extend ExWAS analyses

Supported By

This course is supported by the National Institutes of Health (NIH):

  • National Institute of Environmental Health Sciences (NIEHS): R01ES032470, U24ES036819
  • National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK): R01DK137993