Herpes simplex virus Research
The Use of DNA Microarrays
to
Analyze Gene Expression in HSV infected Cells
The analysis of cellular gene expression in response to
various stimuli has been given tremendous impetus with the recent
development of DNA microarrays.
Applied to the study of large viruses, this technology empowers
us with the ability to carry out rapid, global studies of viral gene
expression and cellular responses to infection under varying conditions
of infection.
We have established a long-term collaborative program with
Professor Peter Ghazal,
Director of the Scottish Genome Centre
at the University of Edinburgh to design and construct
oligonucleotide-based DNA microarrays in order to study global changes
in viral and cellular transcript abundance following infections with
neurotropic herpesviruses,
especially HSV-1. Our current approach is outlined below.
Click image to view enlargement
The chip is made up of triplicate spots 75-mer
oligonucleotides specific probes for
each HSV-1 transcription unit. Because of the fact that a number of HSV
transcript partially overlap each other, sharing 3'-polyadenylation
sites, not all transcripts can be uniquely detected, but siting probes
in the unique 5'-regions of the distal member of overlapping transcript
groups, and the fact that a number of overlapping sets share a common
kinetic class has allowed the resolution of 50 individual transcripts
out of the 70-odd transcripts expressed with appreciable abundance
during productive infection.
Application of DNA microarray analysis to the
study of HSV transcript abundance.
The replication of HSV in infected cells proceeds through several
phases or "waves" of expression of transcripts encoding various viral
functions. This is readily seen by analyzing patterns of viral
transcript abundance in RNA isolated at various times following
infection of cultured cells, such as human fibroblasts (HFF cells). For
such an experiment, poly(A)-containing RNA is isolated, and used to
prime cDNA synthesis using dye-substituted nucleotides as outlined in
the
animated illustration of the microarray
technique.
Each probe is spotted in triplicate on the chips, and the
median value of the three separate hybridization values, less the
background seen from the median values of a number of "blank" (i.e.,
SSC only) spots is determined, along with the standard error (SD) of
the values.
The hybridization values for each probe group can best be
shown as a table (see
some recent publications), or the values for the resolved
transcripts can be plotted with a bar graph (as shown at left) where
the viral transcripts are grouped according to kinetic class.
Patterns of expression 30, 90, 150, 210, 330, and 510 min
after adding virus show the increasing complexity of transcript
abundance, and the fact that first immediate-early, then early, and
finally late transcripts increase markedly in abundance.
View a time lapse animation of the
time-dependant changes in HSV-1 transcript abundance during productive
infection.
Click image for a larger view
The immediate-early and early stages of infection can be
approximated with appropriate drug treatments. Thus, inhibition of de
novo protein synthesis with cycloheximide
leads to the continued expression of immediate-early transcripts as
illustrated below. It is important to note, however, that the lack of
viral protein synthesis under these conditions leads to expression of
much higher levels of immediate-early transcripts than normally seen at
the earliest times of a productive infection because the
down-regulation of immediate-early
transcription by ICP4 protein does not occur.

click image for a larger view
Similarly, blockage of viral DNA replication with a drug such
as Acyclovir (ACV) or phosphonoacetic acid (PAA) will lead to a
situation where only immediate-early and early transcripts are
expressed with normal abundance. These drugs do not
fully inhibit the expression of the leaky-late transcripts. This is
particularly evident when compared with the transcript abundance
patterns seen during infection with a virus mutant that does not
express viral DNA polymerase, and, thus, is completely blocked in DNA
replication.

click for a larger view
One very convenient way to compare changes in transcript
abundance under varying conditions of infection, infection with
metabolic inhibitors or between normal infection and that with a
defined viral mutant is to carry out "scatter analysis". Here, chips
are hybridized with cDNA made to RNA isolated under the conditions to
be compared, and the normalized median values are plotted against each
other in an X-Y scatter plot using excel or other graphical program.
Those transcripts expressed at the same level under both sets of
conditions will lie on or around a straight line, while those whose
levels of abundance are altered under one condition or the other will
fall off that line.

An example of such analysis is shown above for the transcript
abundances seen in cells infected with the DNA polymerase mutant as
compared to treatment of cells infected with wt virus in the
presence of ACV--note that the data are color coded as in the bar
graphs, and that several values demonstrate significant deviation from
a linear correlation.
Such analysis is useful in
studying the effects of different metabolic inhibitors. For example 4%
DMSO and PAA both lead to profound inhibition of virus replication, and
DMSO inhibits viral DNA replication, it is clear that transcript
abundance is markedly different under the two conditions of infection.
This leads naturally to the conclusion that the DMSO-based inhibition
of viral replication is not solely the result of inhibition of viral
DNA replication. In fact, at least three specific viral processes
necessary for normal levels of viral replication are inhibited with
this compound.
Application of Resonance Light Scattering (RLS)
for the Detection of Hybrids
While the approaches outlined above are useful for the study of viral
gene expression in cultured cells, it is clear that significant
increases in sensitivity will enhance the methods applicability for the
study of viral gene expression in animal models. Towards this end,
higher sensitivity detection of hybridized cDNA can be accomplished by
using colloidal gold and silver labeling of cDNA and resonance light
scattering (RLS) measurements. The
Invitrogen
Corporation is marketing through its Genicon subsidiary both
labeled nucleotides and a light scattering scanner. In our laboratory
the colloidal gold and silver labeling protocol along with the use of a
mixing hybridization chamber (
Biomicrosystems)
allows hybridization and ready analysis of samples of 100-200 ng of
poly(A) RNA from infected cells; a 10-fold increase in sensitivity.
Such methods coupled with other amplification methods should allow
hybridization of RNA from a few 10s of cells or less from infected
tissue.
Data Analysis
Statistical Validation of Microarray Data
All HSV probes are spotted in triplicate, and numerical values for
hybridization are obtained by laser scanning of the chip to measure
fluorescence (expressed in arbitrary units) derived from the spotted
oligonucleotide. This value is first adjusted by subtraction of
the background fluorescence of an equivalent area within a concentric
ring just outside the spotted sample, then this value is reduced to a
net value by subtraction of the median of a large number (ca 100)
regions spotted with SSC buffer alone during fabrication. The median of
the three replicate samples and the standard deviation for each probe
is expressed using Microsoft Excel.
With Laser scanners, there are two adjustable variables that
must be set for each individual scan, the laser power and the photo
multiplier gain. While each can be adjusted to any value between
0 and 100%, useful values for the laser power lie between 40 and 90
with the photo multiplier set at some constant value or a set number of
units more or less than laser power. Thus, in practice only one
variable need be set.
The ratio of fluorescent signal to actual sample value is
linear only to net (-SSC) values of 40,000 or so. While theoretically
this allows a greater than 40-fold discrimination in hybridization
values, the fact that weak fluorescent signals are inherently less
reliable than strong ones makes the actual range of reciprocity
considerably smaller. Properly controlled, fluorescent signals
differing by as much as 3 logs can be reliably differentiated--provided
multiple scans are utilized.
In order to compare data from repeat and time-varying
experiments, the chip hybridization data is normalized as
follows. Analysis begins by taking the median signal of probe
values for each transcript set, and the 75th percentile rank for the
total viral hybridization is calculated. One experimental group
in a set of replicates is chosen
as representative, and the values of other groups are adjusted to an
equivalent 75th percentile value. This is simply a means of
emphasizing the higher significance of high hybridization values as
compared to lower ones. An example of replicate data before and
after normalization are shown below.

For publishing and statistical confidence in any findings, the
minimum number of statistically independent samples is three.
This helps to indicate data variation representative of a population
and allows rudimentary hypothesis testing by non-parametric statistical
methods. Obviously, the reliability is greater with more independent
samples.
Hybridization data can be expressed as a ratio between a
control and an experimental determination (as graphically determined by
scatter analysis as described above), as a numerical value related to
fluorescence signal intensity (other examples shown above), or as a
measure of relative abundance based upon the ratio of a given probe
signal to the total viral signal under any given conditions. In
the latter case, it is important to include numerical signal strengths
as a measure of hybridization no mater what the final form of
presentation. Further, knowledge of the background signal seen
with non-specific probes or SSC-spotted blanks provides an index of
general reliability of the hybridization in questions.
Resolution of overlapping transcripts.
Fluorescent-labeled nucleosides are quite bulky, and synthesis of cDNA
using either oligo-dT or random-oligomer priming yields products
averaging much less than 500 bp in length. Thus, 75-mers probes
representing sequences within 300 bp of polyadenylation site of each of
the 50-odd transcription units can be readily detected using oligo-dT
primers; however, oligomers representing sequences upstream of this
region will not hybridize to oligo-dT primed cDNA made with
dye-substituted nucleotides. If polyadenylated RNA is isolated
and cDNA generated by random oligomer primers, however, upstream
oligonucleotide probes can provide data concerning transcripts
represented by them. This is illustrated below for several sets
of overlapping transcripts.

In the experiment shown, RNA was isolated at 4 hr following
infection of HeLa cells--a time at which mainly early transcripts are
abundant. cDNA was synthesized either using an oligo-dT primer
with 20 mg total RNA with Cy5-substituted
dUTP, or with a random hexamer primer with 1 mg
of poly(A)-selected RNA and Cy3-substituted dUTP. The cDNAs were
then hybridized to an HSV DNA microarray, and selected data for the UL27/28,
UL39/40, UL44/45, US5/6/7, and US8/9
overlapping transcript sets. In each case, the oligo-dT primed
cDNA (green) only hybridized to the 3' probe. The random-primed
cDNA (red) provided signals for both probe except with the US5/6/7
transcript group where the distal US5 transcript is abundant
at late times while the US7 is abundant early.