15 In Silico Prediction of the Point of Departure (POD) …
303
15.2.1 An Example from Thomas et al. [17]
Thomas et al. [17] discussed the temporal concordance between apical
and transcriptional PODs for several chemicals. The chemicals are 1,2,4Tribromobenzene (TRBZ), 2,3,4,6-tetrachlorophenol (TTCP), Bromobenzene
(BRBZ), 4,4
-Methylenebis (N,Ndimethyl) benzenamine (MDMB), Hydrazobenzene (HZBZ), and N-Nitrosodiphenylamine (NDPA). A specific strain and sex of
rat as well as route of exposure were chosen for each chemical. Rats were exposed
to each chemical at five dose levels for 5 days and 2, 4, or 13 weeks. Rats were
randomly assigned to each dose groups with ten rats per group. Liver is the target
tissue for TRBZ, BRBZ, TTCP, and HZBZ. Bladder is the target tissue for MDMB
while thyroid is the target tissue for NDPA.
After exposure, the target tissue was harvested for both histological evaluation
and transcriptional profiling. Typically, ten rats were evaluated per concentration per
time point for histological changes. RNA was isolated from six rats per dose per time
points. After purification, RNA from five rats per concentration per time point with
the best quality was used for microarray analysis using the Affymetrix HT RG-230
PM Array Plate.
Transcriptional POD analysis. The transcriptional POD was determined with
the benchmark dose (BMD) approach. Genewise expression levels were analyzed
using BMDExpress (v. 1.41, [18]), which enables automatic model selection and
integration with biological pathways. Here, it is worthwhile to detail the modeling
parameters used in this study while different potential choices will be discussed in
the next subsection. In this study, the microarray data were log 2 transformed and normalized with the Robust Multi-Array Average normalization method (RMA [19]).
The normalized intensity values were then fit into four different dose-response models with BMDExpress: linear, two-degree polynomial, three-degree polynomial, and
power models. The BMD was calculated as the dose where the estimated response
is 1.349 times the standard deviation of the response at dose zero. A statistical lower
bound estimate of a confidence interval for the BMD (BMDL) was also derived. To
select a single model for POD determination, the likelihood ratio test (for nested
models: linear, 2-degree polynomial, 3-degree polynomial) and the Akaike information criterion (AIC) were used. The best fitting model was used to calculate the
BMD and BMDL. To avoid the effect of probe sets with poorly fitting models, it
was further required that the BMD value to be lower than the highest dose and the
goodness-of-fit p-value <0.01.
Once the BMD and BMDL were calculated for each probe set, they were aggregated at the pathway level. Probe sets were mapped to unique genes. Those mapped
to multiple genes were removed from analysis. For genes represented by more than
one probe set, BMDs and BMDLs were averaged to derive values for the gene.
Gene identifiers were matched to pathways using the GeneGo Metacore database.
Pathways with fewer than five genes with BMDs of required quality were removed
from analysis. Median values for BMDs and BMDLs for each pathway were used
as pathway level values.
303
15.2.1 An Example from Thomas et al. [17]
Thomas et al. [17] discussed the temporal concordance between apical
and transcriptional PODs for several chemicals. The chemicals are 1,2,4Tribromobenzene (TRBZ), 2,3,4,6-tetrachlorophenol (TTCP), Bromobenzene
(BRBZ), 4,4
-Methylenebis (N,Ndimethyl) benzenamine (MDMB), Hydrazobenzene (HZBZ), and N-Nitrosodiphenylamine (NDPA). A specific strain and sex of
rat as well as route of exposure were chosen for each chemical. Rats were exposed
to each chemical at five dose levels for 5 days and 2, 4, or 13 weeks. Rats were
randomly assigned to each dose groups with ten rats per group. Liver is the target
tissue for TRBZ, BRBZ, TTCP, and HZBZ. Bladder is the target tissue for MDMB
while thyroid is the target tissue for NDPA.
After exposure, the target tissue was harvested for both histological evaluation
and transcriptional profiling. Typically, ten rats were evaluated per concentration per
time point for histological changes. RNA was isolated from six rats per dose per time
points. After purification, RNA from five rats per concentration per time point with
the best quality was used for microarray analysis using the Affymetrix HT RG-230
PM Array Plate.
Transcriptional POD analysis. The transcriptional POD was determined with
the benchmark dose (BMD) approach. Genewise expression levels were analyzed
using BMDExpress (v. 1.41, [18]), which enables automatic model selection and
integration with biological pathways. Here, it is worthwhile to detail the modeling
parameters used in this study while different potential choices will be discussed in
the next subsection. In this study, the microarray data were log 2 transformed and normalized with the Robust Multi-Array Average normalization method (RMA [19]).
The normalized intensity values were then fit into four different dose-response models with BMDExpress: linear, two-degree polynomial, three-degree polynomial, and
power models. The BMD was calculated as the dose where the estimated response
is 1.349 times the standard deviation of the response at dose zero. A statistical lower
bound estimate of a confidence interval for the BMD (BMDL) was also derived. To
select a single model for POD determination, the likelihood ratio test (for nested
models: linear, 2-degree polynomial, 3-degree polynomial) and the Akaike information criterion (AIC) were used. The best fitting model was used to calculate the
BMD and BMDL. To avoid the effect of probe sets with poorly fitting models, it
was further required that the BMD value to be lower than the highest dose and the
goodness-of-fit p-value <0.01.
Once the BMD and BMDL were calculated for each probe set, they were aggregated at the pathway level. Probe sets were mapped to unique genes. Those mapped
to multiple genes were removed from analysis. For genes represented by more than
one probe set, BMDs and BMDLs were averaged to derive values for the gene.
Gene identifiers were matched to pathways using the GeneGo Metacore database.
Pathways with fewer than five genes with BMDs of required quality were removed
from analysis. Median values for BMDs and BMDLs for each pathway were used
as pathway level values.