Joint analysis of three seemingly

1
Joint analysis of three seemingly independent
microarray experiments via multivariate
mixed-model equations with null residual
covariance structure
Antonio Reverter1, Yong-Hong Wang1, Keren A
Byrne, Siok Hwee Tan1, Gregory S Harper1, Heather
L. Bruce2, and Sigrid A Lehnert1 The Cooperative
Research Centre for Cattle and Beef
Quality 1CSIRO Livestock Industries, Queensland
Bioscience Precinct 306 Carmody Rd, St Lucia, QLD
4067, Australia 2Food Science Australia, Tingalpa
DC, QLD 4173, Australia
ABSTRACT A bovine microarray of muscle and fat
cDNAs was used in three gene expression
experiments (EXP1, EXP2, and EXP3). EXP1 (14
slides) contrasts the gene expression profiles of
muscle in Brahman steers fed varying quality
diets. EXP2 (12 slides) compares the expression
profile in muscle tissue between two breeds. EXP3
(22 slides) studies the mechanisms underlying
adipogenesis in vitro. This study undertook to
jointly analyse these three experiments using
multivariate mixed-models with the relaxed
assumption of a non-zero correlation amongst gene
expressions across experiments while imposing a
null residual covariance structure. Equivalent
editing criteria were
applied across experiments yielding 1,424,322
intensity records analysed by fitting a
tetra-variate model with 263,938 equations and 18
(co)variance components. The latter were
estimated by restricted maximum likelihood.
Correlation estimates for gene expressions ranged
from 0.477 ? 0.004 (for EXP1 and EXP3) to 0.896 ?
0.002 (for EXP1 and EXP2). These moderate to
strong estimates indicate the power to be gained
with the joint analysis. The expected percentage
of differentially expressed genes, as measured
from the gene by treatment interaction, was 1.2
for diet in EXP1, 1.3 for breed in EXP2, and
17.1 and 5.1 for age effect and adipogenic
treatment, respectively in EXP3.
1st
EXP1 Rockhampton Diets (LD Muscle)
Same Microarray Slide Across the 3
Experiments (48 slides)
MEDIUM (4 ? Animals)
LOW (3 ? Anim, 1 Rep)
2nd
(pooled 3 Anim) HIGH (pooled 2 Anim)
CORRELATION EXP1 EXP2 0.84 (7,515) EXP1
EXP3 0.56 (8,717) EXP2 EXP3 0.51
(7,316)
There is gain to be made from a joint analysis
MEDIUM (Pool Ampl)
LOW (Pool Ampl)
MODEL
EXP1 EXP2 EXP3
TotAmpRNA AmpRNA TotRNA
AmpRNA Log(Intensity) Systematic 1,344 1
,152 672 1,440 Gene 9,245 9,245
9,245 9,245 GeneTrt 27,678
45,096 45,474 104,102 Error
405,820 329,247 219,956 469,299
EXP2 Marbling Breeds (LD Muscle)
TIME 1
TIME 2
TIME 3
BREED 1
BREED 1
BREED 1
3rd
REML V(G) 9.3 9.7 3.8
8.8 V(GT) 0.1 0.1 1.5 0.6
V(E) 1.4 1.5 3.3 2.0
BREED 2
BREED 2
BREED 2
4 Variables 1,424,322 Records 263,938
Equations 18 (co)variances
EXP3 Adipogenesis (in vitro)
T 1
Differentially Expressed Total
Variation 1.2 1.3 17.1
5.1 Actual N (?95 CI) 403 354
n/a 546
T 2
T 3
T 4
T 5
4th
RESULTS Jointly Differentially Expressed
T 6
TREAT 1
TREAT 2
Up Down Breed 1 Breed 2 Treat 1
Treat 2 Diet Restriction Up 84 0 2
1 2 1 Down 319 20
5 9 12 Breed 1 121
0 2 2 Breed 2 233 15
2 Treat 1 403
0 Treat 2 143
IMPLICATIONS Go beyond standard
applications Pleiotropic Patterns of
Co-Regulation Evolution Plasticity
5th