Sei sulla pagina 1di 32

Review

Harnessing the power of gene microarrays for the


study of brain aging and Alzheimers disease:
Statistical reliability and functional correlation
E.M. Blalock
a,
*
, K.-C. Chen
a
, A.J. Stromberg
b
, C.M. Norris
a,c
,
I. Kadish
d
, S.D. Kraner
a
, N.M. Porter
a
, P.W. Landeld
a
a
Department of Molecular and Biomedical Pharmacology, University of Kentucky Medical
Center, 800 Rose St. MS-309, Lexington, KY 40536-0084, USA
b
Department of Statistics, University of Kentucky, Lexington, KY, USA
c
Sanders-Brown Center on Aging, University of Kentucky, Lexington, KY, USA
d
Department of Cell Biology, University of Alabama, Birmingham, AL, USA
Received 24 April 2005; accepted 17 June 2005
Abstract
During normal brain aging, numerous alterations develop in the physiology, biochemistry and
structure of neurons and glia. Aging changes occur in most brain regions and, in the hippocampus,
have been linked to declining cognitive performance in both humans and animals. Age-related
changes in hippocampal regions also may be harbingers of more severe decrements to come from
neurodegenerative disorders such as Alzheimers disease (AD). However, unraveling the mechanisms
underlying brain aging, AD and impaired function has been difcult because of the complexity of the
networks that drive these aging-related changes. Gene microarray technology allows massively
parallel analysis of most genes expressed in a tissue, and therefore is an important new research tool
that potentially can provide the investigative power needed to address the complexity of brain aging/
neurodegenerative processes. However, along with this new analytic power, microarrays bring
several major bioinformatics and resource problems that frequently hinder the optimal application of
this technology. In particular, microarray analyses generate extremely large and unwieldy data sets
and are subject to high false positive and false negative rates. Concerns also have been raised
regarding their accuracy and uniformity. Furthermore, microarray analyses can result in long lists of
altered genes, most of which may be difcult to evaluate for functional relevance. These and other
problems have led to some skepticism regarding the reliability and functional usefulness of
microarray data and to a general view that microarray data should be validated by an independent
method. Given recent progress, however, we suggest that the major problem for current microarray
www.elsevier.com/locate/arr
Ageing Research Reviews
4 (2005) 481512
* Corresponding author. Tel.: +1 859 323 8033; fax: +1 859 323 1981.
E-mail address: emblal@uky.edu (E.M. Blalock).
1568-1637/$ see front matter # 2005 Elsevier Ireland Ltd. All rights reserved.
doi:10.1016/j.arr.2005.06.006
research is no longer validity of expression measurements, but rather, the reliability of inferences
from the data, an issue more appropriately redressed by statistical approaches than by validation with
a separate method. If tested using statistically dened criteria for reliability/signicance, microarray
data do not appear a priori to require more independent validation than data obtained by any other
method. In fact, because of added condence from co-regulation, they may require less. In this article
we also discuss our strategy of statistically correlating individual gene expression with biologically
important endpoints designed to address the problem of evaluating functional relevance. We also
review how work by ourselves and others with this powerful technology is leading to new insights
into the complex processes of brain aging and AD, and to novel, more comprehensive models of
aging-related brain change.
# 2005 Elsevier Ireland Ltd. All rights reserved.
Keywords: Bioinformatics; Replication; Hippocampus
1. Promise and pitfalls of gene microarrays
Gene microarrays (Fig. 1) provide a powerful new approach for addressing the
complexity of brain function and aging-related/neurodegenerative processes (Ginsberg
et al., 2000; Miller et al., 2001; Blalock et al., 2003, 2004; Melov and Hubbard, 2004;
Mirnics and Pevsner, 2004) in that they assess the simultaneous activity of thousands of
genes (Schena et al., 1996; Lockhart and Barlow, 2001) and by inference, of multiple
pathways and processes. Thus, microarrays allow investigators to develop more
comprehensive overviews of the interplay among processes and of the context in which
any specic molecule or pathway may be operating. To date, microarrays have been used to
study aging in non-neural tissues of several animal models (Zou et al., 2000; Lee et al.,
2001; Pletcher et al., 2002; Kyng et al., 2003; Landis et al., 2003; Golden and Melov, 2004;
McCarroll et al., 2004; Tollet-Egnell et al., 2004). Moreover, several studies also have
examined aging and/or age-related neurodegenerative changes in the CNS using
mammalian models (Lee et al., 2000; Jiang et al., 2001; Prolla and Mattson, 2001;
Blalock et al., 2003; Dickey et al., 2003; Mirnics et al., 2003, 2005) and human post
mortem Alzheimers disease (AD) tissue (Ginsberg et al., 2000; Loring et al., 2001;
Mufson et al., 2002; Yao et al., 2003; Blalock et al., 2004; Lu et al., 2004) as discussed
further below.
However, the other side of the coin is that microarray analyses are costly and generate
massive amounts of data. These factors lead to several signicant problems, including
difculty in (1) statistically estimating or correcting for error, (2) designing studies with
appropriate statistical power or (3) drawing inferences regarding the functional relevance
of any of the hundreds (or thousands) of expression changes that may be detected. The rst
problem leads to an extensive but often un-quantied proportion of false positives; the
second, to numerous false negatives and high false positive/true positive ratios (false
discovery rate) and the third problem, to potentially long lists of altered genes (in which it
can be very difcult or impossible to determine whether any given change is functionally
relevant, compensatory or largely incidental to the process under study). These problems
have generated considerable skepticism regarding the validity/reliability of microarray
E.M. Blalock et al. / Ageing Research Reviews 4 (2005) 481512 482
data and have led some researchers to question even the general usefulness of microarrays.
In recent years there has been growing recognition of the rst problem (e.g., Miller et al.,
2001) and the resulting increased stringency with which genes are assigned signicant
status has begun to alleviate some of the false positive morass. And important
developments in the area of assessing co-regulation (see below) will ease the third problem
by helping to convert long lists of genes to more interpretable lists of related gene groups
and pathways. Nonetheless, the recognition of these issues, particularly of the second
problem (false negatives) and its interconnectedness with the rst problem (false
positives), is still low and redressing the third problem (functional association) requires
additional approaches.
Fig. 2 illustrates one version from a study on AD of an algorithm we have developed for
addressing some of the major bioinformatics problems faced by microarray studies. In this
algorithm multiple testing error is ameliorated by ltering the data (e.g., excluding
transcripts rated absent or that are unannotated ESTs) so that tests unnecessary to the
E.M. Blalock et al. / Ageing Research Reviews 4 (2005) 481512 483
Fig. 1. Scanned image of a single AD subjects extracted hippocampal RNA hybridized to an HG-U133A
GeneChip
1
oligonucleotide microarray (Affymetrix). The dots of uorescent light are emitted from short
oligonucleotide probes (features) representing unique sequences of genes. The intensity of uorescence activated
by the laser scanner at each probe is proportional to hybridization, and a collection of probes examining different
unique regions of the same RNA sequence (a probe set) is used to determine the expression level of a specic gene
or unannotated EST.
analysis at hand are not performed, thereby reducing expected false positives. In addition,
formal statistical tests are performed for each remaining gene, so that a clear estimate of
maximum expected false positives could be obtained. Importantly, sufciently large
sample sizes (and power) are used in these statistical tests to reduce false negatives and
increase total positives, thereby also reducing the false discovery rate (FDR = expected
false positives/total positives) in the multiple comparisons. Thus, both the expected false
positive rate and false negatives are reduced by these approaches. Further, to address the
problem of assessing functional relevance, we have employed a strategy of statistical
correlation of each tested gene with functional endpoints (e.g., with MMSE and NFT in
AD; Fig. 2), an approach that also benets greatly from large sample sizes and good
statistical power. These and similar approaches, along with the new methods for functional
categorization being developed (see below, Section 3), appear to hold considerable promise
for helping microarray analyses to realize their full potential as powerful tools for gene
discovery. Some of these issues and approaches, as well as their application in studies of
aging/AD, are discussed further in following sections.
E.M. Blalock et al. / Ageing Research Reviews 4 (2005) 481512 484
Fig. 2. Gene identication algorithm redressing some bioinformatics problems in a study of AD. (A) Genes rated
absent (see methods, Blalock et al., 2004) were excluded from analysis. (B) Only annotated probe sets (not
expressed sequence tags) were included in the statistical analysis. (C) Pearson correlation was performed for every
gene against both MiniMental Status Exam (MMSE) and neurobrillary tangle count (NFT) measures of each
subject. Venn diagram shows the number of genes signicantly correlated ( p 0.05) with both MMSE and NFT
or either index alone. For each index, the false discovery rate (FDR) was calculated. (D) For the genes found to
correlate signicantly across all subjects (overall, n = 31), another Pearsons correlation was performed post hoc
among only the subjects rated either Control or Incipient AD (Incipient, n = 16). (Reprinted from
Proceedings of the National Academy of Sciences; Blalock et al., 2004.)
2. Validity versus reliability of microarray data
Most of the analysis and interpretive problems noted above are bioinformatics-related.
However, another perceived problemof microarray analyses has been that they are subject to
variable quality control and technical inaccuracy. Based on a combination of these analytical
and technical concerns, it is often suggested that selected mRNA values obtained by
microarray analysis should be validated using a second measurement technique, such as
quantitative real-time PCR(rtPCR). The rationale for validationappears tohave originated in
the early years of microarray development when nearly all laboratories made their own arrays
and when many different approaches to array manufacture, cDNA probe construction,
mechanical spotting and data analysis were in use (Schena et al., 1996; Lockhart and Barlow,
2001). Further, the published annotations of transcript databases were in many cases
preliminary and/or contradictory. And as noted above, the large data sets and the costs
associated with microarray studies ledthe vast majority of studies torely on inferences froma
handful of microarray chips. Thus, accuracy and reproducibility of results and quality control
became major concerns, stimulating a perceived need for independent validation.
Measurement theory separates a tests validity (does the test measure what it purports to
measure) from its reliability (does the test nd reproducible answers on repeated
applications), and the error in early microarray studies arose from sources involving both
inadequate validity and inadequate reliability. One major source of error, technical
inaccuracy in probe construction that generates gene misidentication or invalid
expression values, clearly is an issue of test validity that can be controlled by validation
with an independent method (e.g., PCR). Asecond error source, however, inadequate inter-
chip reproducibility for the same sample, represents a problem of technical reliability that
is controlled more effectively by rigorous quality control and data on inter-chip testing than
through validation by an independent measurement technique.
In recent years, however, the rapid progress in the genomics and microarray elds has
considerably ameliorated issues of technical validity and reliability, and the rise of large
commercial and academic manufacturers that expend considerable resources on quality
control and database curation has led to much greater uniformity and accuracy in the
annotation of shared transcript databases and chip reproducibility. Studies on inter-chip
reproducibility by large manufacturers such as Affymetrix show that the measurement
accuracy issue is becoming largely non-existent (Affymetrix, 2001; Hollingshead et al.,
2005). Actually, in many cases, microarray analysis seems more accurate than the
independent technique used for validation. For example, a number of laboratories have
found difculties with the accuracy of rtPCR, as minute volumes and exponential
amplication can magnify the slightest pipetting error. Moreover, selecting a single
housekeeping gene for normalization can be more subject to serious pitfalls than
microarray normalization procedures that often utilize much of the chip (Bustin, 2000).
Further, few microarray studies report the statistical parameters for the PCR or other
validation analysis. Clearly, when PCR and microarray data disagree, the question can
become how do we validate the validation? Given these developments and
considerations, therefore, the need for independent validation of microarray data does
not seem any greater than for data obtained with other widely used molecular biology
techniques that rely on manufacturers reagents.
E.M. Blalock et al. / Ageing Research Reviews 4 (2005) 481512 485
2.1. Inferential reliability
Nevertheless, there are some forms of error that are of particular concern to microarray
research. Multiple sources of error arising from bioinformatics problems, including small
sample sizes and/or issues of experimental design/statistics, as noted above, still appear to
pose potentially serious pitfalls for many microarray experiments. This kind of error, which
generates incorrect inferences regarding differential gene expression across groups and
conditions, can occur in the context of fully accurate measurement (validity) and high chip
reproducibility (technical reliability). Therefore, it is important to note that errors of
statistical inference are related to issues of reliability, not validity.
In many prior microarray studies gene changes have been identied largely based on
fold-change magnitude, usually in pair-wise comparisons and in small samples (often
reduced further by pooling-see discussion on pooling in Peng et al., 2003). But false
positives are readily generated when gene identication analyses rely primarily on fold-
change differences in small groups. Fold-change selection depends solely on magnitude of
difference and neglects the sample variance on which formal statistics are based. Under
these conditions, very large magnitude differences frequently can be found simply by
chance and/or by one or both measures in the analysis being extremely small (see
discussions in Miller et al., 2001; Blalock et al., 2003).
However, even if appropriate variance-based statistics are used, a major source of false
positives in microarray studies arises from multiple testing (comparison) error (Miller
et al., 2001; Blalock et al., 2003). That is, testing thousands of genes for signicant change
(e.g., at the p 0.05 level) implies that up to 5% of the genes tested will be found positive
by chance alone (expected false positives). This error arises unavoidably because of the
principle that for each statistical test we perform, there is a small probability (the alpha
level, e.g., p 0.05) that a difference of the magnitude observed will occur by chance
alone. For example, comparing 10,000 genes across two conditions would generate up to
500 expected false positives in the data set of total positives found at p 0.05. If the p-
value is set lower (e.g., p 0.01) more reliable differences will be identied and only 100
false positives will be expected. The downside to the more stringent p-value is that fewer
true positives will be identied, thereby increasing false negative error (Blalock et al.,
2003).
Again, it is important to emphasize that the source of fold-change or multiple testing
error is not the microarray measurements, which may be perfectly accurate. Instead it is the
biological variability of the population, which is such that sample mean differences of a
certain magnitude will occur in 5% of samples (tested at p 0.05) from this population by
chance alone. This has nothing to do with the validity of the analytic procedure, and would
occur with any conventional univariate technique that an investigator repeated numerous
times. Further, if a series of false positive expression changes were found in a set of
microarray samples because of multiple testing error or other statistical problems, re-
analysis of the same RNA samples with an independent, accurate technique would nd the
same false positives. That is, because multiple testing error is due to variability in the
sample, not in the microarray measurement of that sample, the differences in expression it
generates are real and would be conrmed (rather than identied as error) by PCR
validation of the same RNA samples. However, it is much less likely that results from these
E.M. Blalock et al. / Ageing Research Reviews 4 (2005) 481512 486
false positives would be conrmed/replicated if the experiment were repeated on a new set
of samples (in contrast to inferences based on true positives).
Thus, irreproducible inferences (false positives and false negatives), which appear by far
to be the major sources of error and uncertainty in current microarray studies, reect
problems of reliability not validity. Such problems of reliability are most effectively
addressed statistically, for example, either byindependent replication(overlapping positives)
(Miller et al., 2001; Norris et al., 2005) or byincreasinggroupsample size, suchthat statistical
condence is enhanced (Blalock et al., 2003). Consequently, we suggest that rather than
dedicating considerable resources and effort to independent validation, microarray studies
should instead increasingly allocate those resources to enhancing statistical reliability.
3. Co-regulation of functionally related genes
One of the major benets of microarray analysis is that the observations are not limited
to single molecules, but also provide insights into alterations of biochemical pathways or
classes of related genes. This more comprehensive perspective clearly has many
advantages, ranging fromimproved functional interpretation to enhanced condence in co-
regulated positives. Unlike conventional univariate measures (e.g., Western and Northern
blots, rtPCR, in situ hybridization or immunocytochemistry), massively parallel
microarray-based measurements confer the benet of guilt-by-association functional
grouping (Mirnics et al., 2001; Quackenbush, 2003). Researchers can cross-index
signicant results for multiple genes with a priori dened functional groups and biological
pathways, and the alteration of multiple genes in a pathway by the same treatment
strengthens conclusions regarding the role that pathway may play in the phenomenon being
studied. Furthermore, the observation that multiple genes of a certain category or pathway
are over-represented among identied signicant genes in a study lends considerable added
(albeit, quantitatively undened) condence in the result for any one of those genes
(Mirnics et al., 2001; Blalock et al., 2004).
Genes can be assigned individually to pathways or functional categories by an
investigator based on researcher knowledge and review of literature (Jiang et al., 2001;
Prolla and Mattson, 2001; Blalock et al., 2003), or automatically based on databased gene
product functions (Redfern et al., 2000; Blalock et al., 2004; Pass et al., 2004; Pavlidis
et al., 2004). The former technique has the advantage that the researcher probably has a
greater depth of knowledge regarding the function of particular members of the
transcriptome than is contained in a generalized database, but suffers from the
disadvantage that other pathways involvement may be underemphasized or even
unrecognized. Therefore, several research groups have developed and are actively curating
databases for automatic functional assignment, such as the Gene Ontology (GO, Harris
et al., 2004), the Kyoto Encyclopedia for Genes and Genomes (KEGG, Kanehisa et al.,
2004) and Swiss-Prot (Gasteiger et al., 2001; Boeckmann et al., 2003). Additionally, a new
database specically addressing aging-related changes, the Human Aging Genomic
Resources (HAGR), has recently been developed (de Magalhaes et al., 2005).
While it is possible to perform functional grouping statistical analyses on an ad hoc
basis (e.g., Blalock et al., 2003), programs, such as Expression Analysis Systematic
E.M. Blalock et al. / Ageing Research Reviews 4 (2005) 481512 487
Explorer (E.A.S.E., Hosack et al., 2003; Blalock et al., 2004; Norris et al., 2005),
MAPPFinder (Doniger et al., 2003) and Onto-Express (Draghici et al., 2003) are simpler to
use, well-designed, and free to basic researchers. These programs can assign genes to
functional groups within various ontologies and calculate statistics regarding the probability
of particular functional groups being over- or under-represented (Fig. 3). These automated
procedures can cut analysis time from months to hours (Draghici et al., 2003) and can be of
great value to functional interpretation in microarray research (Blalock et al., 2004; Table 1).
Importantly, research is ongoing regarding the weighting of results in pathway analysis.
It is possible that, within a pathway, certain gene changes are more important than others,
and thus perturbations in their expression should be weighted more heavily in conclusions
regarding that pathways involvement in a treatment or condition. Evidence of these
keystone changes can be obtained from the literature and there are several programs and
interactive websites (reviewed in Chaussabel, 2004) that can perform these operations,
including multiplex PubMed queries (Becker et al., 2003) and queries that discover
relationships based on text searches within published documents (LaBaer, 2003; Muller
et al., 2004; Zhou et al., 2004; Alako et al., 2005) (also see commercial applications, e.g.,
Stratagene Pathway Assist (www.stratagene.com)).
E.M. Blalock et al. / Ageing Research Reviews 4 (2005) 481512 488
Fig. 3. Functional categorization (over-representation analysis) of genes identied by microarray analysis. First
level: researchers use a test criterion (e.g., p-value for ANOVA, t-test, correlation analysis, or possibly fold-
change) to identify genes that are signicantly altered across treatments in a microarray study. Second level: the
number of genes found to be signicant is divided by the total number of genes tested on the chip to calculate the
study wide probability of signicant genes. For instance, if a study had 10,000 genes and 1000 of them were found
to be signicant, then the study wide probability would be 10% (1000/10,000). Biological categories can be
dened through databases like GenAge, Gene Ontology, or even manually by researchers. Using the Gene
Ontology as an example, categories (or nodes) within the ontology can be queried as separate gene lists. The
proportion of genes identied as signicant within a particular list is compared to the proportion found in the study
wide probability. Using a statistical procedure, such as Fischers exact test, the binomial test, the hypergeometric
test or the EASE score, researchers can assign a statistical probability to the likelihood that such a proportion
would be likely to occur by chance. For proportions that are greater than the study wide probability (over-
represented), as determined by the statistical test, one can infer that more genes in the category in question are
altered than would be expected by chance alone.
E
.
M
.
B
l
a
l
o
c
k
e
t
a
l
.
/
A
g
e
i
n
g
R
e
s
e
a
r
c
h
R
e
v
i
e
w
s
4
(
2
0
0
5
)
4
8
1

5
1
2
4
8
9
Table 1
Results from an automated biological process over-representation analysis in the AD brain
Up-regulated (1572/6265; 25.1%) EASE N/M/B Down-regulated (1126/6265; 18.0%) EASE N/M/B
Regulation of transcription
(269/792; 34%)
0.0000 21/38/41 Energy pathways (57/151; 37.7%) 0.0000 15/15/69
Cell proliferation (210/666; 31.5%) 0.0001 23/43/35 ATP biosynthesis (16/23; 69.6%) 0.0000 18/9/73
Oncogenesis (24/47; 51.1%) 0.0003 21/39/39 Synaptic transmission (49/143; 34.3%) 0.0000 9/30/61
Protein a.a. phosphorylation
(104/310; 33.5%)
0.0006 23/30/47 Coenzyme biosynthesis (20/40; 50%) 0.0000 15/15/69
Transition metal ion homeostasis
(10/16; 62.5%)
0.0076 18/45/36 Cation transport (60/197; 30.5%) 0.0000 13/18/69
Positive reg. cell prolif.
(25/62; 40.3%)
0.0119 18/68/14 Protein folding (30/86; 34.9%) 0.0003 32/11/57
Establishment. . . chromatin arch.
(34/94; 36.2%)
0.0186 25/43/33 Tricarboxylic acid cycle
(12/22; 54.5%)
0.0006 27/27/47
Nucleosome assembly (13/27; 48.1%) 0.0219 11/56/33 Glycolysis (14/29; 48.3%) 0.0007 6/18/76
Histogenesis and organogenesis
(22/57; 38.6%)
0.0319 22/17/61 Neurogenesis (64/244; 26.2%) 0.0011 19/27/53
Cell adhesion (94/314; 29.9%) 0.0346 19/46/35 Amino acid catabolism (13/30; 43.3%) 0.0038 33/0/67
Development (235/850; 27.6%) 0.0425 21/42/37 Ubiquitin-dep. protein catab.
(27/87; 31%)
0.0043 48/13/39
Complement activation\, classical
(9/18; 50%)
0.0576 10/40/50 Secretion (14/37; 37.8%) 0.0095 0/35/65
Negative reg. cell prolif. (28/83; 33.7%) 0.0762 09/50/41 Protein transport (66/288; 22.9%) 0.0245 26/25/49
Isoprenoid metabolism (6/10; 60%) 0.0789 00/83/17 Neurotransmitter metabolism (6/11; 54.5%) 0.0329 17/17/67
Apoptosis (72/255; 29.5%) 0.0818 13/32/55 Axon guidance (8/19; 42.1%) 0.0404 27/9/64
Defense response (102/360; 28.3%) 0.1010 15/57/28 Calcium ion transport (11/32; 34.4%) 0.0482 7/7/87
Lipid metabolism (82/288; 28.5%) 0.1250 15/47/38 Microtubule-based process (20/73; 27.4%) 0.0538 11/21/68
Biological process categories signicantly over-represented by up-regulated (left) and down-regulated (right) Alzheimers disease-dependent genes (ADGs) ( p < 0.15;
EASE score) are shown. Similar categories are excluded to reduce redundancy. After each category description (in parentheses) is the fraction and percentage of
signicant/total gene associations with that category. The ratios for total identied up- and down-regulated genes are shown in the headings. EASE, modied Fishers
exact test p-value; N/M/B, percentage of genes included in category because they were signicant by NFT correlation (N), MMSE correlation (M) or both (B). (Reprinted
from Proceedings of the National Academy of Sciences; the complete list of ADGs is given alphabetically in web Table 5 of Blalock et al., 2004.)
It is important to note that these increasingly rened approaches for the detection of co-
regulation and of pathway or functional class involvement (e.g., Pavlidis et al., 2004), are
rapidly advancing microarray research from an early developmental stage in which it
primarily yielded unprocessed lists of genes to a newand sophisticated level fromwhich
it can provide insights into dynamic alterations in complex biological systems and
functions. This advancement in co-regulation analysis clearly will be a key element in
redressing the problems of evaluating functional relevance. Additionally, another valuable
approach to functional relevance appears to be that of correlation with functional
endpoints, a strategy that we have implemented extensively as summarized below.
4. Correlations for functional relevance
As discussed, co-regulation provides insights into altered categories/pathways, thereby
aiding functional interpretation. Nonetheless, the functional implications of changes in
many identied genes, or even pathways, may not be apparent in a given study. Thus, one of
the major problems facing microarray studies is howto evaluate the functional relevance of
any specic gene that may be on a long list of genes identied by the analysis. Obviously,
the gold standard for establishing causative relevance is an intervention study. However,
while correlation of course cannot establish causation, what is less well recognized is that
correlation is a key prediction of a causative relationship; that is, two causally related
variables should co-vary. Therefore, since it is not feasible to perform intervention studies
for each gene that may be found to change (particularly in well-powered studies, e.g.,
Blalock et al., 2004), nding genes that change in correlation with key functional endpoints
is a next best strategy (Fig. 4). This approach allows substantial narrowing of potential
candidate genes for intervention studies, as any that are not consistently correlated with
function are much less likely to be directly relevant.
In our studies of aging-related and neurodegenerative processes, we have, as noted, used
large samples of independent subjects, excised regions of interest, and used a separate
microarray chip for each subject, to counter problems associated with multiple testing,
types I and II error, tissue heterogeneity and inter-subject variability. In addition, we have
applied the strategy of correlating microarray changes with behavioral/functional outcome
markers to strengthen the interpretation of transcriptional prole alterations specically
associated with the condition under investigation. In our study of Alzheimers disease in
post mortem tissue (Blalock et al., 2004), for example, we correlated expression of each of
thousands of genes with two established markers of AD, the Mini-Mental Status Exam and
a neurobrillary tangle index (see Fig. 4); and in a study of normal aging in rats (Blalock
et al., 2003), we correlated gene expression with behavioral tests of cognitive function (see
Fig. 1 in Blalock et al., 2003). It is important to emphasize that the processes under study
generally are graded and variable, and consequently, even within a single age group, there
are variations among individuals. In the aging study, some aged animals performed as well
as their younger counterparts, while in the AD study, some subjects with the same clinical
diagnosis varied considerably in pathological markers. These correlative approaches have
allowed identication not simply of genes that change with brain aging or AD, but more
importantly, of those that change in relation to functional decline. Together, the results have
E.M. Blalock et al. / Ageing Research Reviews 4 (2005) 481512 490
provided new clues and generated new models of the functional pathogenesis of AD and
brain aging (see below).
5. Reproducibility of ndings across studies
The next section briey considers and compares the microarray studies performed to
date on aging brain in humans and animals, as well as AD models and human AD brain
tissue. As is apparent from the preceding discussion, microarray studies vary dramatically
in platforms, sample sizes, identication algorithms and statistical approaches, and studies
in the brain aging and AD elds are no exception. Furthermore, studies in this eld vary
E.M. Blalock et al. / Ageing Research Reviews 4 (2005) 481512 491
Fig. 4. Gene identication through correlation. For each gene, expression intensity is plotted on the y-axis, and
MMSE (A left and C left) or neurobrillary tangle (NFT) (B right and D right) scores are plotted on the x-axis; R
2
value, p-value (Pearsons test), linear t (black line) and 95% condence intervals (dashed lines) are also shown.
The MMSE scale is reversed, so that AD severity increases to the right on all x-axes. (A and B) Genes for which
expression levels were up-regulated with AD, identied by negative or positive correlation with MMSE (A) or
NFT (B) scores, respectively. (C and D) Genes for which expression levels were down-regulated with AD,
identied by positive or negative correlation with MMSE (C) or NFT (D), respectively. (Reprinted from
Proceedings of the National Academy of Sciences; Blalock et al., 2004.)
considerably in the models used and the brain regions studied. Given this variability, it
might be expected that the results fromthese studies also vary substantially, and in fact, this
is clearly the case.
As discussed above, reliability (reproducibility) of ndings across studies is a hallmark
of true positives. That is, the probability of the same false positive result occurring by
chance in two separate tests is quite low, reecting the product of p-value probabilities
found on both tests (e.g., 0.05 0.05 = 0.0025). In addition, comparing results from two
similar microarray studies can determine whether the rate of replication (i.e., overlapping
positive results) is greater than might be expected by chance, and therefore, facilitate
interpretation of whether a common underlying process is generating the true positive gene
changes. A greater-than chance rate of replication can be determined, for example, by
the binomial test, comparing the degree of overlap in the two studies to that expected by
chance alone (e.g., Fig. 5). Thus, replication in two tests provides enhanced statistical
condence in scientic inferences.
Nonetheless, failure to repeat a signicant expression change does not necessarily imply
that the gene alteration observed in the rst study was a false positive. There can, of course,
be many other reasons for such discrepancies, including, importantly, a false negative
result in one of the studies. Because of concerns about the numerous false positives
generated by multiple comparisons in microarray studies, investigators often use stringent
criteria for gene identication (i.e., for determining that a genes expression has changed).
While this approach reduces false positives, it conversely increases false negative error.
The latter is compounded by the low group ns commonly employed in microarray studies,
which yield low statistical power with high false negatives. However, even when power is
reasonably high (e.g., 0.7), there is a strong chance of a false negative result and
accordingly, the chance of replicating a true positive effect in two independent studies (e.g.,
0.7 0.7 = 0.49) may not be better than 50%. In addition, cellular heterogeneity, regional
localization or model specicity of changes, microarray platform sensitivity, experimental
design, and a priori assumptions by researchers may all contribute variability and account
for different results among studies.
In light of the many causes of potential disagreements in ndings across studies, as
well as the difculty in evaluating the basis for such disagreements, replication in two
independent studies confers added condence in the overlapping positive results. And
although a smaller number of gene changes agree than disagree between studies to date,
the smaller agreeing set can likely be viewed as a robust and reliable set of biomarkers
for brain aging or AD. Consequently, in the next section, we summarize results and
analyze the overlap of gene changes between our studies and the multiple and diverse
other studies of brain aging and AD. For several reasons, we used our own recent works
as references against which to compare the other studies, including that we have
conducted large and relatively similar microarray studies in hippocampus of both
normally aging rats (Blalock et al., 2003) and AD subjects (Blalock et al., 2004). Our
studies also were well-powered (ns of 29 and 31, respectively), and our total numbers of
signicant genes (348 in aging, 3413 in AD on a larger chip) much exceeded expected
false positives (99 and 967, respectively). From these total numbers of observed
positives, we determined which were also identied in at least one other study of aging
or AD/AD models.
E.M. Blalock et al. / Ageing Research Reviews 4 (2005) 481512 492
The degree of overlap with other studies and the conditions for each study are shown in
Table 2. Note that many genes identied in other studies were not even rated present on our
chips (because of different patterns of expression in different species, platforms and brain
regions as well as varying criteria for absence calls). Note also that for those genes that
were common, the highest proportions of overlap (replication) were found between our
studies and those with the largest sample sizes (even when comparing disparate models),
presumably because of the fewer false negatives expected in higher-powered studies.
Specic genes that overlapped between our studies and at least one other are described
below. Although these are relatively short lists considering the total positives identied, the
genes on the lists can be viewed as including a high proportion of true positives, for reasons
noted above.
E.M. Blalock et al. / Ageing Research Reviews 4 (2005) 481512 493
Fig. 5. Binomial analysis of overlapping positives in two studies. Comparison of genes found to change
(positives) both as a function of aging (Ag) in Blalock et al. (2003) and in response to calcineurin (CaN)
overexpression in Norris et al. (2005). Binomial analysis revealed a degree of overlap of positives signicantly
greater than expected by chance for genes found to agree in direction in the two studies (i.e., calcineurin Up/Ag
Up, red circle; calcineurin Down/Ag Down, green circle). Values for binomial probabilities of deviation from
expected overlap are plotted on the x-axis (converted to standard deviations). Frequencies of the probabilities for
different degrees of deviation are plotted on the y-axis. In contrast, note the relative under-representation of genes
found to disagree in direction in the two studies (i.e., calcineurin Up/Ag Down, pink; calcineurin Down/Ag Up,
blue). The high degree of overlap suggests that aging and CaN may activate related pathways. (Reprinted with
permission from Journal for Neuroscience; Norris et al., 2005.) Similar overlap analyses can be used to compare
two studies on the same subject (e.g., microarray studies from different labs on post mortem ADtissue) in order to
assess statistical reliability of positive results (see text). (For interpretation of the references to colour in this gure
legend, the reader is referred to the web version of the article.)
E
.
M
.
B
l
a
l
o
c
k
e
t
a
l
.
/
A
g
e
i
n
g
R
e
s
e
a
r
c
h
R
e
v
i
e
w
s
4
(
2
0
0
5
)
4
8
1

5
1
2
4
9
4
Table 2
Agreeing/overlapping transcriptional alterations across studies
Reference Species Platform Tissue Design # Genes changed # Also present
in Blalock
et al. (2003)
% Agree
" #
Comparison of aging studies (vs. Blalock et al., 2003)
Blalock et al. (2003) Rat Affymetrix Hip CA1 n = 910/group,
1 array/animal
348 348 (178"/170#)
1. Jiang et al. (2001) Mouse Affymetrix Cortex n = 3/group,
pooled, 1 array/
group
97 38 (201/184) 25 11
2. Lee et al. (2000) Mouse Affymetrix Neocortex n = 3/group, 1
array/animal
186 73 (39"/34#) 31 0
3. Lu et al. (2004) Human Affymetrix Frontal cortex n = 30, 1 array/
subject
320* 119 (64"/55#) 33 7
# Also present
in Blalock
et al. (2004)
Comparison of AD/AD models (vs. Blalock et al., 2004)
Human AD
Blalock et al.
(2004)
Human Affymetrix CA1 tissue n = 31, 1 array
per subject
3414 3413
(1977"/1436#)
4. Colangelo et al.
(2002)
Human Affymetrix CA1 tissue n = 6/group,
pooled, 1 array/group
38 34 (17"/17#) 29 29
5. Ginsberg et al.
(2000)
Human Spotted CA1 neurons n = 20/group, pooled,
4 arrays/group
isolated neurons
32 26 (26#) 35
6. Loring et al.
(2001)
Human Spotted Cing. ctx/amygdala n = 69/group, 1
array/subject
106* 88 (41"/47#) 27 49
7. Mufson et al.
(2002)
Human Nylon Nucleus basalis n = 610/group 12 8 (1"/7#) 100 14
8. Pasinetti (2001) Human Spotted Supratemporal
gyrus
n = 56/group,
pooled, 1 array/group
18 13 (13#) 54
E
.
M
.
B
l
a
l
o
c
k
e
t
a
l
.
/
A
g
e
i
n
g
R
e
s
e
a
r
c
h
R
e
v
i
e
w
s
4
(
2
0
0
5
)
4
8
1

5
1
2
4
9
5
Reference Species Platform Tissue Design # Genes changed # Also present
in Blalock
et al. (2004)
% Agree
" #
9. Yao et al. (2003) Human Affymetrix Superior
frontal gyrus
n = 56/group, 1
array/subject
5 4 (4#) 50
AD models
10. Beglopoulos et al.
(2004)
Mouse Affymetrix Cortex n = 1/group, 1
array/group
56* 34 (21"/13#) 5 8
11. Dickey et al.
(2003)
Mouse Spotted Ctx and hip vs. cblm,
strtm and brstm
n = 4/group, 1
array/animal
43 18 (11"/7#) 100 57
12. Lee et al. (2004) Mouse Spotted Amygdala n = 5/group, pooled,
1 array/group
21 19 (8"/11#) 0 2
13. Mirnics et al.
(2003)
Mouse Affymetrix Brain n = 6/group, 1
array/animal
124* 59 (20"/39#) 70 56
14. Mirnics et al.
(2005)
Mouse Affymetrix Ctx and hip n = 45/group, 1
array/animal
57* 43 (24"/19#) 33 21
15. Reddy et al.
(2004)
Mouse Spotted Cortex n = 5/group,
pooled, 4 arrays/group
240* 169 (63"/106#) 16 24
Platform: Microarray design used in study. Tissue abbreviations: Hip, hippocampus; CA1, cornu ammonis 1 region of hippocampus; cing., cingulate; ctx, cortex; cblm,
cerebellum; strtm, striatum; brstm, brainstem. # Genes changed: number of genes reported to change by authors (* includes only mRNA transcripts annotated with gene
symbols); # also present in Blalock et al. (2003, 2004): the total number reported to change by respective authors that also were rated present and annotated in the Blalock
et al. (2003, 2004) studies (and showing the direction of change indicated in the authors work); % agree: the percentage of total genes found both in the respective
authors work and in the Blalock et al. (2003, 2004) studies that are both signicant (positive) and agree in direction. Boldface indicates a signicantly higher percent of
genes agree between the two studies than would be expected by chance (signicant over-representation, p < 0.05, by the binomial test). Based on the percent of positive
genes found among all tested in Blalock et al. (2003, 2004), the expected probability in the binomial test (Fig. 5) for overlap/agreement was 10% for aging comparisons
with Blalock et al. (2003) (references 13), and 19.9 and 14.5% for up- and down-regulated genes, respectively, in the AD comparisons with Blalock et al. (2004)
(references 415). Note: Many of the AD model mice differ importantly in the nature of their transgenic engineering.
Table 2 (Continued)
6. Overview and overlap of microarray studies on brain aging and aging-related
disease
6.1. Mammalian models of brain aging (Table 3)
As seen in Tables 2 and 3, most microarray studies (as other studies) of brain aging have
been performed in rodent models. Essentially all have found a clear increase in the
expression of genes related to inammatory, oxidative and glial processes (Lee et al., 2000;
Jiang et al., 2001; Blalock et al., 2003). Increases in glial and/or inammatory markers
have been seen in many conventional non-microarray studies as well (Rogers et al., 1996;
Murray and Lynch, 1998; Buttereld et al., 1999; Bickford et al., 2000; Calingasan and
Gibson, 2000; Andreasson et al., 2001; Mrak and Grifn, 2001; Nicolle et al., 2001; Wang
et al., 2001; Finch et al., 2002; Gemma et al., 2002; Wyss-Coray and Mucke, 2002; Wenk
et al., 2003), and therefore, their detection in microarray studies can perhaps be viewed as a
kind of positive control, validating microarray technology. And, in fact, such changes may
represent low hanging fruit for microarray analysis in that up-regulated genes in this
category make up the great majority of identied genes that overlapped across studies
(Table 3). This overlap appears to reect both strong expression and relatively ubiquitous
aging changes across brain regions and models. Interestingly, the highest overlap of our rat
aging study (Blalock et al., 2003) was with the human tissue study of normal aging
(Table 2, Lu et al., 2004), which had similar sample size to ours. Of course, many changes
may not be as strong or ubiquitous as inammatory responses, and therefore, their
detection may require greater biological replication (to increase statistical power) and more
selective brain tissue isolation and preparation (to reduce cellular heterogeneity) (Becker,
2002; Ginsberg et al., 2000).
In our microarray-based study on brain aging in a rat model (Blalock et al., 2003), we
focused on the CA1 region of the hippocampus, an area long implicated in aging-related
cognitive decline (Landeld and Pitler, 1984; Disterhoft et al., 1993; Thibault et al.,
1998). Each animals CA1 region was hybridized to a separate microarray and three age
groups (young, mid-aged and aged) were used-to assess (roughly) at which point along
the age-course transcriptional alterations began to occur. Further, each animals
performance on two hippocampally dependent tasks (Spatial Water Maze and Object
Memory Task) was assessed prior to euthanasia. Analysis was performed using formal
statistical tests after results had been ltered for absence calls and non-annotated
transcripts. Absence call ltering resulted in a marked improvement in the expected
false discovery rate of our data set. This is not unexpected as the expression values for
probe sets rated absent by the Microarray Suite algorithms (MAS4 and MAS5) are
considered by the manufacturer to be too volatile for analysis (Affymetrix, 2001).
Transcriptional changes were identied at two levels; rst, by whether the genes
expression differed signicantly across age (tested with one-way ANOVA) and second,
in a subset, by whether it also correlated (by Pearsons test) signicantly with cognitive
performance. Functional grouping analysis was performed using literature searches and
supported through automated Gene Ontology assignments. Genes that met criteria for
both aging dependence and functional correlation were identied as aging and cognition
related genes (ACRGs).
E.M. Blalock et al. / Ageing Research Reviews 4 (2005) 481512 496
As noted, our studies were in agreement with many others regarding altered expression
of molecules related to inammatory responses, oxidative stress, glial activation and
mitochondrial dysfunction. Many of these molecules are reected in Table 3. In addition,
this well-powered study yielded many ACRGs that were not detected in other studies. From
E.M. Blalock et al. / Ageing Research Reviews 4 (2005) 481512 497
Table 3
Agreement (overlap) of gene transcription changes in the brain with aging
Gene References
Up-regulated with age (Blalock et al., 2003)
Apod Apolipoprotein D 2,3
Cast Calpastatin 3
Cd9 CD9 antigen 2
Cgef2 cAMP-regulated guanine nucleotide exchange factor II 3
Cryab Crystallin, alpha B 2
Csrpi Cysteine and glycine-rich protein 1 3
Ctsd Cathepsin D 2
Dck Deoxycytidine kinase 1
Ddr1 Discoidin domain receptor family, member 1 3
Erbb3 V-erb-b2 erythroblastic leukemia viral oncogene homolog 3 3
Fah Fumarylacetoacetate hydrolase 1
Fgfr2 Fibroblast growth factor receptor 2 3
Gatm Glycine amidinotransferase 3
Gfap Glial brillary acidic protein 2,3
Gna15 Guanine nucleotide binding protein, alpha 15 2
Hbb Hemoglobin beta chain complex 3
Hcrtr2 Hypocretin (orexin) receptor 2 3
Klk6 Kallikrein 6 1
Lamp2 Lysosomal membrane glycoprotein 2 3
Lcat Lecithin cholesterol acyltransferase 2
Mag Myelin-associated glycoprotein 2
Mog Myelin oligodendrocyte glycoprotein 2
Msn Moesin 3
Nfe2l2 NF-E2-related factor 2 3
Pmp22 Peripheral myelin protein 22 3
S100b S100 protein, beta polypeptide 3
Seppi Selenoprotein P, plasma, 1 1,3
Tgfbr2 Transforming growth factor, beta receptor II 3
Tmp21 Integral membrane protein Tmp21-I (p23) 3
Vamp1vesi Vesicle-associated membrane protein (synaptobrevin 2) 2
Vegf Vascular endothelial growth factor form 3 2,3
Vim Vimentin 1,2,3
Down-regulated with age (Blalock et al., 2003)
Ap2m1 Adaptor-related protein complex 2, mu 1 subunit 1
Ca4 Carbonic anhydrase 4 3
Calb1calbi Calbindin 1 3
Coro1a Coronin-1A 3
Gad1 Glutamate decarboxylase 1 3
Pfn1 Prolin 1 1
Genes that changed in both Blalock et al. (2003) and at least one other mammalian brain aging study. References:
references with similar ndings: 1, Jiang et al. (2001); 2, Lee et al. (2000); 3, Lu et al. (2004).
the categories of these ACRGs we were able to identify several novel processes not
previously closely associated with brain aging, such as down-regulated structural synaptic
plasticity, activity regulated signaling and transcription factors, as well as up-regulated
myelin turnover, cholesterol synthesis/transport, protein processing and some signal
transduction, among others. These results, combined with the prior results fromour lab and
others, led us to suggest a new, more comprehensive model of functional brain aging
(Fig. 6), in which altered Ca
2+
signaling leads to decreased neuronal signaling, in turn
reducing biosynthesis and metabolism in neurons, down-regulating synaptic plasticity and
inducing axonal regression. Axonal regression then triggers demyelination, producing
myelin fragments that activate microglia to induce an inammatory response. In parallel,
activated microglia exacerbate the demyelinating response by attacking oligodendrocytes,
which respond with remyelinating programs (e.g., cholesterol synthesis) in a compensatory
fashion. Further, microglia cause a feed-forward increase in inammatory response,
increasing antigen presentation and causing altered glial metabolism, astrocytic
hypertrophy and destabilized extracellular matrix.
This model of course, is only one of several possible cascades that might t the data.
However, it illustrates how the broader overviews of dynamic changes in multiple
pathways now being provided by microarray research are beginning to generate more
comprehensive and presumably, more realistic models for experimental testing. Thus, the
smaller set of gene changes replicated in other studies (Table 3) provided evidence of
reliability, whereas the larger set of gene changes not found in other studies (but whose
reliability is nevertheless based on statistical analysis in a well-powered study) provided
new insights necessary for a novel comprehensive model. Notably, we have recently
replicated many of our ndings on novel ACRGs in another well-powered study of
hippocampal aging in rats (in preparation).
6.2. Human AD (Table 4)
Many of the same principles apply in studies of AD as in the aging studies summarized
above. AD has been remarkably resistant to analysis at the molecular level, possibly
because of the high degree of complex interactions that may exist between cell-types in the
affected brain tissue. While several microarray studies of post mortem AD brain tissue
have been conducted, it is important to note the limitations that differences in experimental
design may have on interpretation. Studies can be divided into three basic designs; those
comparing (1) AD-affected versus AD-resistant brain tissues within subjects, (2) AD-
affected brain structures across AD patients and age-matched controls (Pasinetti, 2001;
Colangelo et al., 2002; Mufson et al., 2002; Yao et al., 2003) or (3) some combination of
affected and unaffected tissue across AD patients and age-matched controls (Ginsberg
et al., 2000; Loring et al., 2001). The within-subject design is powerful because single
subjects can serve as their own controls, helping to reduce the contribution of inter-subject
variability. However, results in such designs may also be inuenced by changes unique to
a potential interaction between the effects of AD and a specic brain region. In addition,
relatively few studies (Pasinetti, 2001; Blalock et al., 2004) have looked at early stages of
ADwith microarrays. Because these early-stage changes presumably are more subtle, it is
likely that such studies need good statistical power (achieved through increased n and/or
E.M. Blalock et al. / Ageing Research Reviews 4 (2005) 481512 498
E.M. Blalock et al. / Ageing Research Reviews 4 (2005) 481512 499
Fig. 6. Integrative model of brain aging. Numbers represent one putative sequence of events consistent with
ndings on aging-dependent hippocampal genomic changes correlated with cognitive dysfunction. Arrows
indicate hypothesized causal interactions for the inammatory cascade component. Altered Ca
2+
and synaptic
signaling (1) in neurons (N) reduce neural activity responses, which then activate genomic alterations that down-
regulate activity-dependent signaling pathways (2) and induce general neuronal, metabolic and biosynthetic
involution (3a and b). These involutional changes induce other transcriptional alterations that down-regulate the
capacity for neurite outgrowth, synaptogenesis, and maintenance of extracellular structure (4). The weakening of
extracellular structure and axonal regression trigger an initial demyelination process (5) that in turn activates
remyelinating programs and associated cholesterol biosynthesis/transport (6a) in oligodendrocytes (O). Con-
currently, myelin fragments are endocytosed by glia and degraded to antigenic epitopes that stimulate innate
autoimmunity and antigen presentation (6b) in microglia (M). These autoimmune responses then activate a glial-
mediated inammatory cascade (7) in microglia (M) and astrocytes (A), associated with altered glial metabolism
(8a) and glucose uptake from capillaries (C), and astrocytic hypertrophy (8b). The increasing inammatory and
glial activation induce additional extracellular matrix transformation and neuronal erosion (9) and exacerbate
demyelination. The accumulating inammatory damage (7) and extracellular changes (9) eventually interact with
decreased neuronal activity (1) and synaptic plasticity (4) to impair cognition and increase neuronal vulnerability
(bottom). (Reprinted with permission from Journal of Neuroscience; Blalock et al., 2003.)
decreased variability) to facilitate detection of incipient changes (e.g., Blalock et al.,
2004).
To identify both early and severe AD changes, we selected a correlative algorithm and a
single subject per array design and correlated each genes expression against a
neuropathologic and a cognitive index of AD pathology. The CA1 region was excised
for microarray analysis from all AD or control subjects (through the Alzheimers Disease
Research Center at the University of Kentucky, Sanders-Brown Center on Aging), each of
whom had received a battery of cognitive tests, including the Mini-Mental Status Exam
(MMSE), and had received post mortem quantication scores for amyloid plaque, diffuse
amyloid and neurobrillary tangle (NFT) density. We used MMSE as a measure of cognitive
function and NFTas a measure of pathology against which each gene was correlated across
subjects (the steps of the full microarray analysis algorithm are summarized in Fig. 2).
Despite the important differences in analytical and design approaches (in addition to the
other important differences between studies, such as array platform and pooling issues),
comparisons of these studies to our own (Tables 2 and 4) again revealed overlapping sets of
genes that can be interpreted as containing a relatively higher proportion of true positives. It
should be noted that, for purposes of comparisons across studies (see below), we used our
results for the changes seen overall across all stages of AD, rather than the smaller set of
results for changes found in the incipient/control correlations (because most other AD
microarray studies examined cases of advanced rather than incipient changes).
Interestingly, in contrast to the normal aging study (Table 3), there was greater agreement
among down-regulated than among up-regulated transcripts (Table 4), including overlap
among many transcripts whose protein products are associated with neuronal function,
suggesting that genomic changes reecting neuronal decline may be a common and clear
hallmark of severe AD. As expected from the large differences between study designs, and
from the aging study, many more gene expression changes in our AD study did not overlap
than agreed with other microarray studies (Tables 2 and 4).
As noted, in this study we additionally identied a subset of transcriptional changes that
were also correlated with AD markers in the very early or incipient stages of AD. Changes
in the incipient stages of AD could provide important clues on the pathogenesis of the
disorder, as well as serve as valuable biomarkers or targets for clinical intervention as new
therapeutic agents are developed. Gene expression correlates of incipient AD revealed an
intriguing alteration in the hippocampal transcriptional prole, including an up-regulation
of messages functioning as transcription factors and tumor suppressors (Blalock et al.,
2004). Based on these ndings, we proposed a model in which alterations in axons or their
myelin sheaths initially stimulate growth and remyelination responses in localized
oligodendrocytes. The oligodendrocytes, in turn, secrete growth factors that have local
autocrine/paracrine stimulatory effects, stimulating oligodendrocytic progenitor cells and
also causing compensatory cell-specic tumor suppressor responses in neurons and
astrocytes. These compensatory responses lead to protein aggregation, affect axonal
myelin interactions, and result in NFTs. As the NFT density increases, wider extracellular
matrix, amyloid precursor protein and inammatory changes may be triggered that impact
cognition. This model could help to explain why AD pathogenesis appears to travel
along myelinated axons fromthe entorhinal cortex to hippocampus and neocortex (Hyman,
1997; Braak and Braak, 1998), leaving NFTs and plaques in its wake.
E.M. Blalock et al. / Ageing Research Reviews 4 (2005) 481512 500
E.M. Blalock et al. / Ageing Research Reviews 4 (2005) 481512 501
Table 4
Agreement (overlap) of gene transcription changes in the human brain with AD
Symbol References
Up-regulated with AD (Blalock et al., 2004)
Add3 Adducin 3 (gamma) 6
Ccl20 Chemokine (CC motif) ligand 20 4
Cdk2Ap1 Cdk2-associated protein 1 6
Cebpd CCAAT/enhancer binding protein (C/EBP), delta 4
Ftl Ferritin, light polypeptide 6
Gak Cycling associated kinase 6
Gja1 Gap junction protein, alpha 1, 43 kDa (connexin 43) 6
Gsta1 Glutathione S-transferase A1 6
Igfbp5 Insulin-like growth factor binding protein 5 6
Il1A Interleukin 1A 4
Nfkb1 Nuclear factor of kappa light polypeptide in B-cells 1 (p105) 4
Pabc1 Poly(A) binding protein, cytoplasmic 1 6
Thg-1 TSC-22-like 6
Tjp2 Tight junction protein 2 6
Tnfaip2 Tumor necrosis factor, alpha-induced protein 2 4
Tob2 Transducer of ErbB-2 6
Tpd52 Tumor protein D52 6
Down-regulated with AD (Blalock et al., 2004)
Chl1 Cell adhesion molecule with homology to L1CAM 6
Cox4L1 Cytochrome c oxidase subunit IV isoform 1 6
Cspg5 Chondroitin sulfate proteoglycan 5 (neuroglycan C) 4
Ddx1 Dead box 1 6
Dscr1L1 Down syndrome critical region gene 1-like 1 6
Ensa Alpha endosulne 6
Fkbp4 FK506 binding protein 4 6
Glrx Glutaredoxin (thioltransferase) 5
Got1 Glutamic-oxaloacetic transaminase 1, soluble 6,8
Gria1 Glutamate receptor, ionotropic, AMPA 1 5
Gria2 Glutamate receptor, ionotropic, AMPA 2 5
6Itpr1 Inositol triphosphate receptor, type 1 6
Kcnb2 Shaker K+ Channel, beta 2 6
Lmo4 Lim domain only 4 6
Lss Lanosterol synthase 6
Map2K1 Mitogen activated protein kinase kinase 1 6
6Mapk1 Mitogen activated protein kinase 1 5,6
Mlf2 Myeloid leukemia factor 2 6
Mdh1 Malate dehydrogenase 1, NAD (soluble) 8
Nars Asparaginyl-tRNA synthetase 6
Ne Neurolament light chain; neuron specic structural protein 4
Nell2 Nel (chicken)-line 2 8
Oxct 3-Oxoacid CoA transferase 6,8
Pgrmd Progesterone receptor membrane component 1 6
Ppp2Ca Protein phosphatase 2, catalytic subunit alpha 5,6
Psma5 Proteasome (prosome, macropain) subunit, alpha type, 5 6
6Psmd1 Proteasome (prosome, macropain) 26S subunit, non-ATPase, 1 6
Rpip8 Rap2 interacting protein 8 6
Rtn1 Reticulon 1 8
Slc25A6 Mitochondrial ADP/ADT translocator, member 6 6
6.3. Animal models of AD (Table 5)
We also compared gene expression proles in our study against those reported in
studies of brain tissue in selected AD models. Initially, we expected to nd a relatively
small degree of overlap considering that the AD models only simulate some aspects of
the human condition. Surprisingly, more genes showed overlapping responses (Tables 2
and 5) than in the human comparison (Table 4). Despite this overlap, there obviously are
many notable differences between human AD and the mouse models including type and
presence of genetic manipulation, time and severity of onset of pathologic and/or
cognitive changes, and species-related differences. Thus, the aspects of gene expression
proles that overlap may be particularly robust. In this section we discuss signicant
overlaps between our work in human AD brain (Blalock et al., 2004) and studies using
animal models of AD.
Double transgenic presenilin1 (PS1)/amyloid precursor protein (APP) mice (Borchelt
et al., 1997) show many characteristics similar to those seen in the human AD population,
such as cognitive decits (Richards et al., 2003), age-dependent accumulation of
pathological markers (Holcomb et al., 1998), and a similar distribution of markers across
hippocampus and cortex (Gordon et al., 2002). Microarray analyses of amyloid bearing
(versus amyloid free) regions of the brains of PS1/APP mice (Dickey et al., 2003) found
up-regulation that overlapped with human AD (Table 5) for inammatory markers (C4a/
C4b and Hspca), lysozomal enzymes (Npc1 and Ctss), and transcription machinery/
regulators (Madh5, Sp100, Rrm2 and Sfrs10), as well as for Cbs (involved in homocysteine
transulferation) and Gadd45B, a negative regulator of cell growth. Interestingly,
overlapping down-regulation was found exclusively for synaptic/intracellular signaling
molecules (Camk2a, Clstn1, Gria1 and Syn47).
Mutated presenilins have been implicated in some forms of Familial Alzheimers
disease. However, the relative activity of these mutated enzymes is not well understood and
may involve both increased and decreased function in different pathways and brain
structures (reviewed in Mirnics et al., 2005). Using a model in which normally functioning
PS1 is knocked out, Mirnics et al. (2003) assessed possible loss of function transcriptional
E.M. Blalock et al. / Ageing Research Reviews 4 (2005) 481512 502
Table 4 (Continued )
Symbol References
Slit2 Slit homolog 2 (Drosophila) 4
Snca Synuclein, alpha (non A4 component of amyloid precursor) 5
6Stat1 Signal transducer and activator of transcription 1, 91 kDa 4
Stx1A Syntaxin 1A (brain) 9
Syn1 Synapsin I 4,5
Syn2 Synapsin II 8
Syt1 Synaptotagmin I 5,7,9
Tpi1 Triosephosphate isomerase 1 6
Tubb2 Tubulin, beta, 2 5
Uchl1 Ubiquitin carboxyl terminal esterase L1 8
Genes that changed in Blalock et al. (2004) and at least one other human AD post mortem brain tissue study.
References: references with similar ndings: 4, Colangelo et al. (2002); 5, Ginsberg et al. (2000); 6, Loring et al.
(2001); 7, Mufson et al. (2002); 8, Pasinetti (2001); 9, Yao et al. (2003).
E
.
M
.
B
l
a
l
o
c
k
e
t
a
l
.
/
A
g
e
i
n
g
R
e
s
e
a
r
c
h
R
e
v
i
e
w
s
4
(
2
0
0
5
)
4
8
1

5
1
2
5
0
3
Table 5
Agreement (overlap) of gene transcription changes between AD and selected AD models
Symbol References Symbol References
Up-regulated with AD (Blalock et al., 2004) Down-regulated with AD (Blalock et al., 2004)
Aebp1 AE binding protein 1 14 Aldoc Aldolase C,
fructose-bisphosphate
15
Appbp2 Amyloid beta precursor
protein binding protein 2
15 Ap2M1 Adaptor-related protein
complex 2, mu 1 subunit
15
Birc4 Baculoviral IAP
repeat-containing 4
13 Arf1 ADP-ribosylation factor 1 15
Bub1 BUB1 budding uninhibited
by benzimidazoles 1 homolog
13 Atp6Ip1 ATPase, H + transporting,
lysosomal interacting protein 1
15
C4A Complement component 4A 11 Atp6V0C ATPase, H + transporting,
lysosomal 16 kDa, V0 subunit C
15
C4B Complement component 4B 11 Atp6V1C1 ATPase, H + transporting,
V1 subunit C, isoform 1
15
Ca1 Carbonic anhydrase I 13 Bat3 HLA-B associated transcript 3 15
Cbs Cystathionine-beta-synthase 11 Bsg Basigin (OK blood group) 15
Cd68 CD68 antigen 14 Camk2A Ca
2+
/CaM dependent
protein kinase Iia
11
Cdc2L2 Cell division cycle 2-like 2 15 Cap Adenylyl cyclase-associated
protein
14
Csh1 Chorionic somatomammotropin
hormone 1
15 Cct3 Chaperonin containing TCP1,
subunit 3 (gamma)
15
Ctss Cathepsin S 11 Chl1 Cell adhesion molecule with
homology toL1CAM
13
Eno1 Enolase 1 alpha 13 Ckb Creatine kinase, brain 15
Epm2A Epilepsy, progressive
myoclonus type 2, Lafora
disease (laforin)
14 Clcn3 Chloride channel 3 13
Exo1 Exonuclease 1 15 Clstn1 Calsyntenin 1 11
Fzd9 Frizzled homolog 9
(Drosophila)
14 Cox4I1 Cytochrome c oxidase
subunit IV isoform 1
15
Gadd45B Growth arrest and
DNA-damage-inducible, beta
10,11 Cox7A2L Cytochrome c oxidase
subunit VIIa polypeptide 2 like
15
E
.
M
.
B
l
a
l
o
c
k
e
t
a
l
.
/
A
g
e
i
n
g
R
e
s
e
a
r
c
h
R
e
v
i
e
w
s
4
(
2
0
0
5
)
4
8
1

5
1
2
5
0
4 Table 5 (Continued )
Symbol References Symbol References
Gfap Glial brillary acidic protein 10,13 Crnkl1 Crn, crooked neck-like
1 (Drosophila)
12
Gja1 Gap junction protein, alpha 1,
43 kDa (connexin 43)
13 Cs Citrate synthase 15
Gpi Glucose phosphate isomerase 10 Dpysl4 Dihydropyrimidinase-like 4 13
Gpx5 Glutathione peroxidase 5 13 Gad1 Glutamate decarboxylase
1 (brain, 67 kDa)
13
Hmox1 Heme oxygenase (decycling) 1 15 Glul Glutamate-ammonia ligase
(glutamine synthase)
15
Hspca Heat shock 90 kDa
protein 1, alpha
11 Gnb1 G protein, beta polypeptide 1 15
Itgb5 Integrin, beta 5 15 Gng3 G protein gamma 3 15
Jag1 Jagged 1 (Alagille syndrome) 14 Got2 Glutamic-oxaloacetictransaminase
2, mitochondrial
15
Lats1 LATS, large tumor suppressor,
homolog 1 (Drosophila)
13 Gria1 Glutamate receptor, ionotropic,
AMPA 1
11
Madh5 MAD, mothers against
decapentaplegic homolog 5
11 Hyou1 Hypoxia up-regulated 1 14
Man2B1 Mannosidase, alpha,
class 2B, member 1
13 Ina Internexin neuronal intermediate
lament protein, alpha
13
Mkrn1 Makorin, ring nger protein, 1 10 Kcnab2 Potassium voltage-gated channel,
shaker beta 2
10
Mtf1 Metal-regulatory transcription
factor 1
15 Kl Klotho 12
Nap1L1 Nucleosome assembly
protein 1, like 1
15 Lxn Latexin protein 13
Npc1 Niemann-pick disease,
type C1
11 Mapre3 Microtubule-associated protein,
RP/EB family 3
13
Npc2 Niemann-pick disease, type C2 13 Mapt Microtubule-associated protein tau 13
Nr4A1 Nuclear receptor subfamily 4,
group A, member 1
14 Meg3 Maternally expressed 3 13
Plagl1 Pleiomorphic adenoma gene-like 1 14 Mgst3 Microsomal glutathione
S-transferase 3
13
E
.
M
.
B
l
a
l
o
c
k
e
t
a
l
.
/
A
g
e
i
n
g
R
e
s
e
a
r
c
h
R
e
v
i
e
w
s
4
(
2
0
0
5
)
4
8
1

5
1
2
5
0
5
Table 5 (Continued )
Symbol References Symbol References
Prelp Proline arginine-rich end
leucine-rich repeat protein
15 Ne Neurolament, light polypeptide
68 kDa
13
Prkcl1 Protein kinase C-like 1 13 Oaz2 Ornithine decarboxylase
antizyme 2
15
Pura Purine-rich element binding
protein A
15 Pcp4 Purkinjecell protein 4 13
Rrm2 Ribonucleotide reductase
M2 polypeptide
11 Pea 15 Phosphoprotein enriched in
astrocytes 15
13,15
Sdc1 Syndecan 1 13 Pias1 Protein inhibitor of activated
STAT, 1
15
Serpini2 Neuroserpin 10 Pnoc Prepronociceptin 13
Sfrs10 Splicing factor,
arginine/serine-rich 10
11 Psa Phosphoserine aminotransferase 13
Sp100 Nuclear antigen Sp100 11 Psmb4 Proteasome subunit, beta 4 15
Sspn Sarcospan
(Kras oncogene-associated gene)
10 Ptprz1 Protein tyrosine phosphatase,
receptor-type Z1
13
Tf Transferring 13 Ptx3 Pentaxin-related gene, rapidly
induced by IL-1 beta
13
Thra Thyroid hormone
receptor, alpha
14 Sgne1 Secretory granule, neuroendocrine
protein 1 (7B2 protein)
14
Tnxb Tenascin XB 13 Slc6A1 Neurotransmitter transporter,
GABA
13
Snap25 Synaptosomal-associated
protein, 25 kDa
13
Sncg Synuclein, gamma 13
Stmn3 Stathmin-like 3 15
Syn47 Homer, neuronal immediate
early gene, 1B
11
Tal1 T-cell acute lymphocytic
leukemia 1
13
E
.
M
.
B
l
a
l
o
c
k
e
t
a
l
.
/
A
g
e
i
n
g
R
e
s
e
a
r
c
h
R
e
v
i
e
w
s
4
(
2
0
0
5
)
4
8
1

5
1
2
5
0
6
Table 5 (Continued )
Symbol References Symbol References
Tm4Sf2 Transmembrane 4 superfamily
member 2
13
Trpc1 Transient receptor potential cation
channel, subfamily C, member 1
14
Ube2H Ubiquitin-conjugating enzyme E2H 15
Uqcrc1 Ubiquinol-cytochrome c reductase
core protein I
15
Ywhaz Tyrosine 3-monooxygenase
act. prot. zeta
15
Genes that changed in both Blalock et al. (2004) and at least one microarray study of brain tissue in an animal model of AD. References: references with similar ndings:
10, Beglopoulos et al. (2004); 11, Dickey et al. (2003); 12, Lee et al. (2004); 13, Mirnics et al. (2003); 14, Mirnics et al. (2005); 15, Reddy et al. (2004).
alterations associated with PS1 ablation and found up-regulation overlapping with human
AD (Table 5) for expression of: negative regulators of cell growth (Birc4, Bub1, Eno1,
Lats1, Tf and Tnxb), intra- and inter-cellular signaling molecules (Gja1 and Prkcl1),
lysozomal enzymes (Man2b1 and Npc2), Ca1, a carbonic anhydrase whose dysfunction is
thought to play a role in AD and aging (reviewed in Sun and Alkon, 2002), Gfap, a marker
for astroglia, Gpx5, a glutathione peroxidase with anti-oxidant properties and Sdc1, a cell-
surface proteoglycan important for cytoskeletal structure. These authors also found
overlapping down-regulation for expression of: neuronal signaling molecules (Clcn3,
Gad1, Pnoc, Psa, Slc6a1 and Snap25), extracellular matrix/cytoskeletal elements (Chl1,
Ina, Mapre3, Mapt, the principle protein product associated with NFTs, Ne and Sncg),
negative regulators of inammation (Pea15, also decreased in APP transgenic mice, Ptx3,
Tal1 and Tm4sf2), regulators of neuronal growth (Dpysl4 and Ptprz1), Lxn, a
carboxypeptidase inhibitor that also serves as a marker of excitatory neurons in cortex
(Job and Tan, 2003), Meg3, an untranslated RNA that may function directly as a tumor
suppressor, Mgst3, which also functions as a glutathione peroxidase and may have anti-
oxidant properties and Pcp4 (Pep-19), which may play a role in sequestering calcineurin
(Slemmon et al., 2000).
APP transgenic mice do not show the neuronal cell loss seen in AD (Stein and Johnson,
2002), but do demonstrate decits in hippocampal-dependent behavioral tasks (Westerman
et al., 2002). Reddy et al. (2004) used microarrays to examine transcriptional proles in the
frontal cortex of APP transgenic mice. Interestingly, signicantly overlapping changes
with human AD (Table 5) were found only for down-regulated genes, including those
related to energy transduction (Aldoc, Ckb, Cox4i1, Cox7a2l, Cs, Got2, Oaz2 and Uqcrc1),
G-protein signaling (Gnb1 and Gng3), protein degradation (Psmb4 and Ube2h), vesicle/
golgi trafcking (Ap2m1, Arf1, Atp6ip1, Atp6V0C and Atp6V1C1), Bat3, a protein of
unknown function coded from the MHC region of the genome, Bsg, a type I membrane
protein whose activity promotes both neurite formation and outgrowth of astrocytic
processes, Cct3, a molecular chaperone, Glul, an enzyme that catalyzes the reversible
conversion of glutamate to glutamine, Pea15, a negative regulator of inammation (also
decreased in PS1 knock out study), Pias1, a transcriptional co-regulator, Stmn3, one of the
neuronal growth associated proteins that may serve as a marker for synaptic plasticity
(Mori and Morii, 2002) and Ywhaz, an adaptor protein that participates in multiple
signaling pathways.
Overall, many of these changes, particularly down-regulated genes associated with
neuron structure and synaptic activity, suggest disorganization of the synaptic structure of
extant neurons or loss of neurons (in the case of PS1 knockouts and APP/PS1 transgenic
animals). The dysfunction in the neuronal compartment is apparently accompanied by glial
activation and inammatory processes, as these markers were also replicated in mouse
models and human AD. Many of these changes also have been observed using non-
microarray based technology and, therefore, the overlapping ndings reviewed here further
strengthen use of animal models to approximate many aspects of AD. Finally, as mentioned
in Section 5, statistically signicant overlapping results do not provide a list of the only true
positive genes that may be common across studies, but rather serve as a stochastic
bellwether, indicating that, at least in part, two phenomena share similar transcriptional
proles.
E.M. Blalock et al. / Ageing Research Reviews 4 (2005) 481512 507
7. Summary and conclusions
In this article, we have considered some of the critical aspects of bioinformatics
problems facing studies employing microarray technology. We also summarized a multi-
pronged strategy we have developed to address, in part, some of these problems. This
strategy comprises the management of multiple comparisons via pre-test ltering, the
application of formal statistical testing of each transcript, use of well-powered sample
sizes to reduce false negatives and improve false discovery rate for identied transcripts,
as well as a one-subject-per-chip experimental design, and the statistical correlation of
individual expression changes with functional endpoints (Blalock et al., 2003, 2004). In
addition, we have briey reviewed the overlap and differences among some recent
ndings on brain aging/AD obtained by ourselves and others using microarray
technology. Together, these results suggest that recent advances in harnessing the raw
analytic power of microarray technology are ushering in a new research era in which it
will be possible to develop more realistic and comprehensive models of dynamically
complex processes, such as aging, disease and systems-level brain functions, than have
previously been feasible.
References
Affymetrix, 2001. Affymetrix Microarray Suite Users Guide, v. 5. Affymetrix, Santa Clara, CA.
Alako, B.T., Veldhoven, A., van Baal, S., Jelier, R., Verhoeven, S., Rullmann, T., Polman, J., Jenster, G., 2005.
CoPub Mapper: mining MEDLINE based on search term co-publication. BMC Bioinform. 6, 51.
Andreasson, K.I., Savonenko, A., Vidensky, S., Goellner, J.J., Zhang, Y., Shaffer, A., Kaufmann, W.E., Worley,
P.F., Isakson, P., Markowska, A.L., 2001. Age-dependent cognitive decits and neuronal apoptosis in
cyclooxygenase-2 transgenic mice. J. Neurosci. 21, 81988209.
Becker, K.G., 2002. Deciphering the gene expression prole of long-lived snell mice. Sci. Aging Knowl. Environ.
2002, pe4.
Becker, K.G., Hosack, D.A., Dennis Jr., G., Lempicki, R.A., Bright, T.J., Cheadle, C., Engel, J., 2003. PubMatrix:
a tool for multiplex literature mining. BMC Bioinform. 4, 61.
Beglopoulos, V., Sun, X., Saura, C.A., Lemere, C.A., Kim, R.D., Shen, J., 2004. Reduced beta-amyloid production
and increased inammatory responses in presenilin conditional knock-out mice. J. Biol. Chem. 279, 46907
46914.
Bickford, P.C., Gould, T., Briederick, L., Chadman, K., Pollock, A., Young, D., Shukitt-Hale, B., Joseph, J., 2000.
Antioxidant-rich diets improve cerebellar physiology and motor learning in aged rats. Brain Res. 866, 211
217.
Blalock, E.M., Chen, K.C., Sharrow, K., Herman, J.P., Porter, N.M., Foster, T.C., Landeld, P.W., 2003. Gene
microarrays in hippocampal aging: statistical proling identies novel processes correlated with cognitive
impairment. J. Neurosci. 23, 38073819.
Blalock, E.M., Geddes, J.W., Chen, K.C., Porter, N.M., Markesbery, W.R., Landeld, P.W., 2004. Incipient
Alzheimers disease: microarray correlation analyses reveal major transcriptional and tumor suppressor
responses. Proc. Natl. Acad. Sci. U.S.A. 101, 21732178.
Boeckmann, B., Bairoch, A., Apweiler, R., Blatter, M.C., Estreicher, A., Gasteiger, E., Martin, M.J., Michoud, K.,
ODonovan, C., Phan, I., Pilbout, S., Schneider, M., 2003. The SWISS-PROT protein knowledgebase and its
supplement TrEMBL in 2003. Nucleic Acids Res. 31, 365370.
Borchelt, D.R., Ratovitski, T., van Lare, J., Lee, M.K., Gonzales, V., Jenkins, N.A., Copeland, N.G., Price, D.L.,
Sisodia, S.S., 1997. Accelerated amyloid deposition in the brains of transgenic mice coexpressing mutant
presenilin 1 and amyloid precursor proteins. Neuron 19, 939945.
E.M. Blalock et al. / Ageing Research Reviews 4 (2005) 481512 508
Braak, H., Braak, E., 1998. Evolution of neuronal changes in the course of Alzheimers disease. J. Neural Transm.
Suppl. 53, 127140.
Bustin, S.A., 2000. Absolute quantication of mRNA using real-time reverse transcription polymerase chain
reaction assays. J. Mol. Endocrinol. 25, 169193.
Buttereld, D.A., Howard, B., Yatin, S., Koppal, T., Drake, J., Hensley, K., Aksenov, M., Aksenova, M.,
Subramaniam, R., Varadarajan, S., Harris-White, M.E., Pedigo Jr., N.W., Carney, J.M., 1999. Elevated
oxidative stress in models of normal brain aging and Alzheimers disease. Life Sci. 65, 18831892.
Calingasan, N.Y., Gibson, G.E., 2000. Dietary restriction attenuates the neuronal loss, induction of heme
oxygenase-1 and blood-brain barrier breakdown induced by impaired oxidative metabolism. Brain Res.
885, 6269.
Chaussabel, D., 2004. Biomedical literature mining: challenges and solutions in the omics era. Am. J.
Pharmacogenomics 4, 383393.
Colangelo, V., Schurr, J., Ball, M.J., Pelaez, R.P., Bazan, N.G., Lukiw, W.J., 2002. Gene expression proling of
12633 genes in Alzheimer hippocampal CA1: transcription and neurotrophic factor down-regulation and up-
regulation of apoptotic and pro-inammatory signaling. J. Neurosci. Res. 70, 462473.
de Magalhaes, J.P., Costa, J., Toussaint, O., 2005. HAGR: the Human Ageing Genomic Resources. Nucleic Acids
Res. 33, D537D543 (Database Issue).
Dickey, C.A., Loring, J.F., Montgomery, J., Gordon, M.N., Eastman, P.S., Morgan, D., 2003. Selectively reduced
expression of synaptic plasticity-related genes in amyloid precursor protein + presenilin-1 transgenic mice. J.
Neurosci. 23, 52195226.
Disterhoft, J.F., Moyer Jr., J.R., Thompson, L.T., Kowalska, M., 1993. Functional aspects of calcium-channel
modulation. Clin. Neuropharmacol. 16 (Suppl. 1), S12S24.
Doniger, S.W., Salomonis, N., Dahlquist, K.D., Vranizan, K., Lawlor, S.C., Conklin, B.R., 2003. MAPPFinder:
using Gene Ontology and GenMAPP to create a global gene-expression prole frommicroarray data. Genome
Biol. 4, R7.
Draghici, S., Khatri, P., Bhavsar, P., Shah, A., Krawetz, S.A., Tainsky, M.A., 2003. Onto-tools, the toolkit of the
modern biologist: onto-express, onto-compare, onto-design and onto-translate. Nucleic Acids Res. 31, 3775
3781.
Finch, C.E., Morgan, T., Rozovsky, I., Xie, Z., Weindruch, R., Prolla, T., 2002. Microglia and aging in the brain. In:
WJ, S. (Ed.), Microglia in the Degenerating and Regenerating CNS. Springer-Verlag, New York.
Gasteiger, E., Jung, E., Bairoch, A., 2001. SWISS-PROT: connecting biomolecular knowledge via a protein
database. Curr. Issues Mol. Biol. 3, 4755.
Gemma, C., Mesches, M.H., Sepesi, B., Choo, K., Holmes, D.B., Bickford, P.C., 2002. Diets enriched in foods
with high antioxidant activity reverse age-induced decreases in cerebellar beta-adrenergic function and
increases in proinammatory cytokines. J. Neurosci. 22, 61146120.
Ginsberg, S.D., Hemby, S.E., Lee, V.M., Eberwine, J.H., Trojanowski, J.Q., 2000. Expression prole of transcripts
in Alzheimers disease tangle-bearing CA1 neurons. Ann. Neurol. 48, 7787.
Golden, T.R., Melov, S., 2004. Microarray analysis of gene expression with age in individual nematodes. Aging
Cell 3, 111124.
Gordon, M.N., Holcomb, L.A., Jantzen, P.T., DiCarlo, G., Wilcock, D., Boyett, K.W., Connor, K., Melachrino, J.,
OCallaghan, J.P., Morgan, D., 2002. Time course of the development of Alzheimer-like pathology in the
doubly transgenic PS1 + APP mouse. Exp. Neurol. 173, 183195.
Harris, M.A., Clark, J., Ireland, A., Lomax, J., Ashburner, M., Foulger, R., Eilbeck, K., Lewis, S., Marshall, B.,
Mungall, C., Richter, J., Rubin, G.M., Blake, J.A., Bult, C., Dolan, M., Drabkin, H., Eppig, J.T., Hill, D.P.,
Ni, L., Ringwald, M., Balakrishnan, R., Cherry, J.M., Christie, K.R., Costanzo, M.C., Dwight, S.S., Engel,
S., Fisk, D.G., Hirschman, J.E., Hong, E.L., Nash, R.S., Sethuraman, A., Theesfeld, C.L., Botstein, D.,
Dolinski, K., Feierbach, B., Berardini, T., Mundodi, S., Rhee, S.Y., Apweiler, R., Barrell, D., Camon, E.,
Dimmer, E., Lee, V., Chisholm, R., Gaudet, P., Kibbe, W., Kishore, R., Schwarz, E.M., Sternberg, P.,
Gwinn, M., Hannick, L., Wortman, J., Berriman, M., Wood, V., de la Cruz, N., Tonellato, P., Jaiswal, P.,
Seigfried, T., White, R., 2004. The Gene Ontology (GO) database and informatics resource. Nucleic Acids
Res. 32, D258D261.
Holcomb, L., Gordon, M.N., McGowan, E., Yu, X., Benkovic, S., Jantzen, P., Wright, K., Saad, I., Mueller, R.,
Morgan, D., Sanders, S., Zehr, C., OCampo, K., Hardy, J., Prada, C.M., Eckman, C., Younkin, S., Hsiao, K.,
E.M. Blalock et al. / Ageing Research Reviews 4 (2005) 481512 509
Duff, K., 1998. Accelerated Alzheimer-type phenotype in transgenic mice carrying both mutant amyloid
precursor protein and presenilin 1 transgenes. Nat. Med. 4, 97100.
Hollingshead, D., Lewis, D.A., Mirnics, K., 2005. Platform inuence on DNA microarray data in postmortem
brain research. Neurobiol. Dis. 18, 649655.
Hosack, D.A., Dennis Jr., G., Sherman, B.T., Lane, H.C., Lempicki, R.A., 2003. Identifying biological themes
within lists of genes with EASE. Genome Biol. 4, R70.
Hyman, B.T., 1997. The neuropathological diagnosis of Alzheimers disease: clinical-pathological studies.
Neurobiol. Aging 18, S27S32.
Jiang, C.H., Tsien, J.Z., Schultz, P.G., Hu, Y., 2001. The effects of aging on gene expression in the hypothalamus
and cortex of mice. Proc. Natl. Acad. Sci. U.S.A. 98, 19301934.
Job, C., Tan, S.S., 2003. Constructing the mammalian neocortex: the role of intrinsic factors. Dev. Biol. 257, 221
232.
Kanehisa, M., Goto, S., Kawashima, S., Okuno, Y., Hattori, M., 2004. The KEGG resource for deciphering the
genome. Nucleic Acids Res. 32, D277D280.
Kyng, K.J., May, A., Kolvraa, S., Bohr, V.A., 2003. Gene expression proling in Werner syndrome closely
resembles that of normal aging. Proc. Natl. Acad. Sci. U.S.A. 100, 1225912264.
LaBaer, J., 2003. Mining the literature and large datasets. Nat. Biotechnol. 21, 976977.
Landeld, P.W., Pitler, T.A., 1984. Prolonged Ca
2+
-dependent afterhyperpolarizations in hippocampal neurons of
aged rats. Science 226, 10891092.
Landis, G.N., Bhole, D., Tower, J., 2003. A search for doxycycline-dependent mutations that increase Drosophila
melanogaster life span identies the VhaSFD, Sugar baby, lamin, fwd and Cctl genes. Genome Biol. 4, R8.
Lee, C.K., Weindruch, R., Prolla, T.A., 2000. Gene-expression prole of the ageing brain in mice. Nat. Genet. 25,
294297.
Lee, H.M., Greeley Jr., G.H., Englander, E.W., 2001. Age-associated changes in gene expression patterns in the
duodenum and colon of rats. Mech. Ageing Dev. 122, 355371.
Lee, K.W., Lee, S.H., Kim, H., Song, J.S., Yang, S.D., Paik, S.G., Han, P.L., 2004. Progressive cognitive
impairment and anxiety induction in the absence of plaque deposition in C57BL/6 inbred mice expressing
transgenic amyloid precursor protein. J. Neurosci. Res. 76, 572580.
Lockhart, D.J., Barlow, C., 2001. Expressing whats on your mind: DNA arrays and the brain. Nat. Rev. Neurosci.
2, 6368.
Loring, J.F., Wen, X., Lee, J.M., Seilhamer, J., Somogyi, R., 2001. A gene expression prole of Alzheimers
disease. DNA Cell Biol. 20, 683695.
Lu, T., Pan, Y., Kao, S.Y., Li, C., Kohane, I., Chan, J., Yankner, B.A., 2004. Gene regulation and DNA damage in
the ageing human brain. Nature 429, 883891.
McCarroll, S.A., Murphy, C.T., Zou, S., Pletche, r S.D., Chin, C.S., Jan, Y.N., Kenyon, C., Bargmann, C.I., Li, H.,
2004. Comparing genomic expression patterns across species identies shared transcriptional prole in aging.
Nat. Genet. 36, 197204.
Melov, S., Hubbard, A., 2004. Microarrays as a tool to investigate the biology of aging: a retrospective and a look
to the future. Sci. Aging Knowl. Environ. 2004, re7.
Miller, R.A., Galecki, A., Shmookler-Reis, R.J., 2001. Interpretation, design, and analysis of gene array
expression experiments. J. Gerontol. A: Biol. Sci. Med. Sci. 56, B52B57.
Mirnics, K., Middleton, F.A., Lewis, D.A., Levitt, P., 2001. Analysis of complex brain disorders with gene
expression microarrays: schizophrenia as a disease of the synapse. Trends Neurosci. 24, 479486.
Mirnics, K., Pevsner, J., 2004. Progress in the use of microarray technology to study the neurobiology of disease.
Nat. Neurosci. 7, 434439.
Mirnics, K., Korade, Z., Arion, D., Lazarov, O., Unger, T., Macioce, M., Sabatini, M., Terrano, D., Douglass, K.C.,
Schor, N.F., Sisodia, S.S., 2005. Presenilin-1-dependent transcriptome changes. J. Neurosci. 25, 1571
1578.
Mirnics, Z.K., Mirnics, K., Terrano, D., Lewis, D.A., Sisodia, S.S., Schor, N.F., 2003. DNAmicroarray proling of
developing PS1-decient mouse brain reveals complex and coregulated expression changes. Mol. Psychiatry
8, 863878.
Mori, N., Morii, H., 2002. SCG10-related neuronal growth-associated proteins in neural development, plasticity,
degeneration, and aging. J. Neurosci. Res. 70, 264273.
E.M. Blalock et al. / Ageing Research Reviews 4 (2005) 481512 510
Mrak, R.E., Grifn, W.S., 2001. Interleukin-1, neuroinammation, and Alzheimers disease. Neurobiol. Aging 22,
903908.
Mufson, E.J., Counts, S.E., Ginsberg, S.D., 2002. Gene expression proles of cholinergic nucleus basalis neurons
in Alzheimers disease. Neurochem. Res. 27, 10351048.
Muller, H.M., Kenny, E.E., Sternberg, P.W., 2004. Textpresso: an ontology-based information retrieval and
extraction system for biological literature. PLoS Biol. 2, e309.
Murray, C.A., Lynch, M.A., 1998. Evidence that increased hippocampal expression of the cytokine interleukin-1
beta is a common trigger for age- and stress-induced impairments in long-term potentiation. J. Neurosci. 18,
29742981.
Nicolle, M.M., Gonzalez, J., Sugaya, K., Baskerville, K.A., Bryan, D., Lund, K., Gallagher, M., McKinney, M.,
2001. Signatures of hippocampal oxidative stress in aged spatial learning-impaired rodents. Neuroscience 107,
415431.
Norris, C.M., Kadish, I., Blalock, E.M., Chen, K.C., Thibault, V., Porter, N.M., Landeld, P.W., Kraner, S.D.,
2005. Calcineurin triggers reactive/inammatory processes in astrocytes and is upregulated in aging and
Alzheimers models. J. Neurosci. 25, 46494658.
Pasinetti, G.M., 2001. Use of cDNA microarray in the search for molecular markers involved in the onset of
Alzheimers disease dementia. J. Neurosci. Res. 65, 471476.
Pass, H.I., Liu, Z., Wali, A., Bueno, R., Land, S., Lott, D., Siddiq, F., Lonardo, F., Carbone, M., Draghici, S., 2004.
Gene expression proles predict survival and progression of pleural mesothelioma. Clin. Cancer Res. 10, 849
859.
Pavlidis, P., Qin, J., Arango, V., Mann, J.J., Sibille, E., 2004. Using the gene ontology for microarray data mining: a
comparison of methods and application to age effects in human prefrontal cortex. Neurochem. Res. 29, 1213
1222.
Peng, X., Wood, C.L., Blalock, E.M., Chen, K.C., Landeld, P.W., Stromberg, A.J., 2003. Statistical implications
of pooling RNA samples for microarray experiments. BMC Bioinform. 4, 26.
Pletcher, S.D., Macdonald, S.J., Marguerie, R., Certa, U., Stearns, S.C., Goldstein, D.B., Partridge, L., 2002.
Genome-wide transcript proles in aging and calorically restricted Drosophila melanogaster. Curr. Biol. 12,
712723.
Prolla, T.A., Mattson, M.P., 2001. Molecular mechanisms of brain aging and neurodegenerative disorders: lessons
from dietary restriction. Trends Neurosci. 24, S21S31.
Quackenbush, J., 2003. Genomics. Microarraysguilt by association. Science 302, 240241.
Reddy, P.H., McWeeney, S., Park, B.S., Manczak, M., Gutala, R.V., Partovi, D., Jung, Y., Yau, V., Searles, R.,
Mori, M., Quinn, J., 2004. Gene expression proles of transcripts in amyloid precursor protein transgenic
mice: up-regulation of mitochondrial metabolism and apoptotic genes is an early cellular change in
Alzheimers disease. Hum. Mol. Genet. 13, 12251240.
Redfern, C.H., Degtyarev, M.Y., Kwa, A.T., Salomonis, N., Cotte, N., Nanevicz, T., Fidelman, N., Desai, K.,
Vranizan, K., Lee, E.K., Coward, P., Shah, N., Warrington, J.A., Fishman, G.I., Bernstein, D., Baker, A.J.,
Conklin, B.R., 2000. Conditional expression of a Gi-coupled receptor causes ventricular conduction delay and
a lethal cardiomyopathy. Proc. Natl. Acad. Sci. U.S.A. 97, 48264831.
Richards, J.G., Higgins, G.A., Ouagazzal, A.M., Ozmen, L., Kew, J.N., Bohrmann, B., Malherbe, P., Brockhaus,
M., Loetscher, H., Czech, C., Huber, G., Bluethmann, H., Jacobsen, H., Kemp, J.A., 2003. PS2APP transgenic
mice, coexpressing hPS2mut and hAPPswe, show age-related cognitive decits associated with discrete brain
amyloid deposition and inammation. J. Neurosci. 23, 89899003.
Rogers, J., Webster, S., Lue, L.F., Brachova, L., Civin, W.H., Emmerling, M., Shivers, B., Walker, D., McGeer, P.,
1996. Inammation and Alzheimers disease pathogenesis. Neurobiol. Aging 17, 681686.
Schena, M., Shalon, D., Heller, R., Chai, A., Brown, P.O., Davis, R.W., 1996. Parallel human genome analysis:
microarray-based expression monitoring of 1000 genes. Proc. Natl. Acad. Sci. U.S.A. 93, 1061410619.
Slemmon, J.R., Feng, B., Erhardt, J.A., 2000. Small proteins that modulate calmodulin-dependent signal
transduction: effects of PEP-19, neuromodulin, and neurogranin on enzyme activation and cellular home-
ostasis. Mol. Neurobiol. 22, 99113.
Stein, T.D., Johnson, J.A., 2002. Lack of neurodegeneration in transgenic mice overexpressing mutant amyloid
precursor protein is associated with increased levels of transthyretin and the activation of cell survival
pathways. J. Neurosci. 22, 73807388.
E.M. Blalock et al. / Ageing Research Reviews 4 (2005) 481512 511
Sun, M.K., Alkon, D.L., 2002. Carbonic anhydrase gating of attention: memory therapy and enhancement. Trends
Pharmacol. Sci. 23, 8389.
Thibault, O., Porter, N.M., Chen, K.C., Blalock, E.M., Kaminker, P.G., Clodfelter, G.V., Brewer, L.D., Landeld,
P.W., 1998. Calciumdysregulation in neuronal aging and Alzheimers disease: history and newdirections. Cell
Calcium 24, 417433.
Tollet-Egnell, P., Parini, P., Stahlberg, N., Lonnstedt, I., Lee, N.H., Rudling, M., Flores-Morales, A., Norstedt, G.,
2004. Growth hormone-mediated alteration of fuel metabolism in the aged rat as determined from transcript
proles. Physiol. Genom. 16, 261267.
Wang, J., Green, P.S., Simpkins, J.W., 2001. Estradiol protects against ATP depletion, mitochondrial membrane
potential decline and the generation of reactive oxygen species induced by 3-nitroproprionic acid in SK-N-SH
human neuroblastoma cells. J. Neurochem. 77, 804811.
Wenk, G.L., McGann, K., Hauss-Wegrzyniak, B., Rosi, S., 2003. The toxicity of tumor necrosis factor-alpha upon
cholinergic neurons within the nucleus basalis and the role of norepinephrine in the regulation of inamma-
tion: implications for Alzheimers disease. Neuroscience 121, 719729.
Westerman, M.A., Cooper-Blacketer, D., Mariash, A., Kotilinek, L., Kawarabayashi, T., Younkin, L.H., Carlson,
G.A., Younkin, S.G., Ashe, K.H., 2002. The relationship between Abeta and memory in the Tg2576 mouse
model of Alzheimers disease. J. Neurosci. 22, 18581867.
Wyss-Coray, T., Mucke, L., 2002. Inammation in neurodegenerative diseasea double-edged sword. Neuron 35,
419432.
Yao, P.J., Zhu, M., Pyun, E.I., Brooks, A.I., Therianos, S., Meyers, V.E., Coleman, P.D., 2003. Defects in
expression of genes related to synaptic vesicle trafcking in frontal cortex of Alzheimers disease. Neurobiol.
Dis. 12, 97109.
Zhou, G., Wen, X., Liu, H., Schlicht, M.J., Hessner, M.J., Tonellato, P.J., Datta, M.W., 2004. B.E.A.R. GeneInfo: a
tool for identifying gene-related biomedical publications through user modiable queries. BMC Bioinform. 5,
46.
Zou, S., Meadows, S., Sharp, L., Jan, L.Y., Jan, Y.N., 2000. Genome-wide study of aging and oxidative stress
response in Drosophila melanogaster. Proc. Natl. Acad. Sci. U.S.A. 97, 1372613731.
E.M. Blalock et al. / Ageing Research Reviews 4 (2005) 481512 512

Potrebbero piacerti anche