Multilevel Models: Geneenvironment Interaction Using TwinSibling Data

1
Multilevel Models Gene-environment
InteractionUsing Twin/Sibling Data

• Guang Guo
• Department of Sociology
• University of North Carolina at Chapel Hill

2
Research on Genes and Behaviorbefore the DNA Era

• Twin and other family studies
• Animal models

3
OFA App.- John Gene
4
(No Transcript)
5
(No Transcript)
6
Hard to do this kind experiments to Humans!
7
Advances in molecular HUMAN genetics

• Before early 1980s, gene research only done by
  expression cloning based on biochemical defects
  and altered proteins.
• A key breakthrough Recognition that the
  variations in human DNA sequence can be used as
  markers for genetic mapping (relate DNA
  variations to diseases and traits).

8

• This led to the mapping of thousands of
  chromosomal regions containing genes affecting
  human diseases.
• Then a variety of molecular genetic techniques
  were used to identify the genes (more than 1,200
  counting).
• Examples include genes for Huntington disease,
  cystic fibrosis, breast cancer (BRCA1, BRCA2),

9

• Mendelian traits (near one-to-one mapping between
  phenotype and genotype) are identified routinely.
• Current focus on complex traits.
• Major efforts planned(Francis Collins, Nature
  2004) prospective study of 200,000 persons with
  a large number of markers (500,000 SNPs) for each
  individual to identify genes of moderate effects
  for 40 common diseases.

10
Gene-Environment Interactions

• -- Caspi et al. Science, 2002 Childhood abuse
  and later criminal activities. Only maltreated
  children with genotype A tended to develop
  behavior problems. Maltreated children with
  genotype B were less affected.
• --Guo Stearns, Social Forces, 2002 Genetic
  potential for cognitive development depends on
  social environment. Poverty depresses expression
  of genes.

11
In this talk, present Mixed or Multilevel Models
for Analysis of Gene-Environment Interaction
Using Twin/Sib Data
Will not talk about genetic analysis using DNA
data today
12
DNA measures would be desirable but for complex
traits, twin/sib analysis is still useful.

• A large number of genes behind (e.g.,more than
  200 for obesity)

13
Logic of Twin Studies

• Identical twins
• A single zygote genetic clones sharing 100
  of the genes
• Fraternal twins Two separate zygotes -- on
  average, have 50 of the genes in common, (just
  like two full siblings).

14
Logic of Twin Studies

• If a trait is shared more often by identical
  twins than by fraternal twins, then genes might
  have played a role in developing the trait.
• This argument holds regardless of number of genes
  involved or whether the alleles involved are
  dominant or recessive.

15
Traditional Twin Analysis

• Data (IQ)

16
Sibling Analysis Method
Shared environment
17

• Unconditional models ( no covariate predictor)

Level-1 model
Level-2 model
18
(No Transcript)
19
Within-pair correlation among MZ and DZ twinsData
20
Results(MZ) covariance parameters
21

• Unlikely the usual case, we are primarily
  interested in the random parameters.

22
Adding observed covariates t is type of pairs
(mz, dz, full sibs) so that different parameters
are estimated for mz, dz, and full sibs
23
(1) SAS codes for model (4) PROC MIXED NOCLPRINT
NOITPRINT COVTEST CLASS CLUSTER_ID TYPE MODEL
PPVTX1 X2 / SOLUTION RANDOM INTERCEPT /
SUBJECTCLUSTER_ID GROUPTYPE REPEATED /
GROUPTYPE RUN
24
Adding random coefficients to the environmental
variables
25
Written as a multilevel model
26
(2) SAS codes for model (7) PROC MIXED NOCLPRINT
NOITPRINT COVTEST CLASS CLUSTER_ID TYPE MODEL
PPVTX1 X2 / SOLUTION RANDOM INTERCEPT /
SUBJECTCLUSTER_ID GROUPTYPE RANDOM X1 X2 /
SUBJECTCLUSTER_ID REPEATED / GROUPTYPE These
SAS codes are the same as previous ones except
the new line five that asks SAS to
estimate random slopes for X1 and X2 at the
cluster level.
27
Calculate correlation coefficients and
then Heritability as a function of environmental
factors
28
An Example of GE interaction

• Adolescents genetic potential for cognitive
  development may be expressed more or less
  depending on social environment.
• Add Health data-Waves I-II
• Outcome variable A version of PPVT
• More than 3000 MZ twins, DZ twins, full sibs, and
  half sibs

29
Reasoning G by E interaction educational outcomes
30
(No Transcript)
31
(No Transcript)
32
(No Transcript)
33
(No Transcript)
34
(No Transcript)
35
(No Transcript)
36
(No Transcript)
37
Conclusion of the Example

• Heritability estimates in a more disadvantaged
  social environment tend to be smaller than that
  in a more normal social environment.

38
Could extend this model longitudinally

• Understand how environmental and genetic factors
  interact to influence behaviors at different ages
  (longitudinal questions)

39
Example

• The National Longitudinal Study of Adolescent
  Health (Add Health)
• MZ and DZ twins, and full siblings (3,000-4,500)
• Three waves 1994-5, 1995-6, and 2002
• Obesity (BMI) 1994-5 (reported) 1995-6 and 2002
  (measured)

40
--- 8 pairs of siblings --- Each person measured
three times (Wave 1, Wave 2, Wave 3)
41
In scalar form j pair or cluster i person t
time period
42
Age 12 - 26
43
Tentative Interpretation

• It suggests that genetic factors play a decreased
  role in young adulthood than in adolescence
  because of more intervention?