Significant Association of Multiple Human Cytomegalovirus Genomic
Loci with Glioblastoma Multiforme Samples
Padhma Ranganathan,a,bPaul A. Clark,eJohn S. Kuo,c,d,e,fM. Shahriar Salamat,gand Robert F. Kalejtaa,b,c,d
Institute for Molecular Virology,aMcArdle Laboratory for Cancer Research,bCarbone Comprehensive Cancer Center,cStem Cell and Regenerative Medicine Center,dand
Departments of Neurological Surgery,eHuman Oncology,fand Pathology and Laboratory Medicine,gUniversity of Wisconsin—Madison, Madison, Wisconsin, USA
counts for nearly half of all central nervous system (CNS) malig-
nancies in adults. Equivalent to a grade IV diffuse astrocytoma,
GBMs consist primarily of neoplastic astrocytes, the most abun-
plastic cell types, including neurons, oligodendrocytes, macro-
phages, and glial and neural stem cells (42). The heterogeneous
nature of these tumors may at least partially explain why they are
refractory to current therapeutics and has led to the hypothesis
although exposure to ionizing radiation, electrical, or magnetic
fields has been proposed as a risk factor (22). Recently, several
reports have detected a potential association between GBMs and
human cytomegalovirus (HCMV), a common betaherpesvirus.
Viruses are causative agents of human cancers (21). At least
15% of all human tumors have a viral etiology. Human cancer
human T-lymphotropic virus type 1 (HTLV-1), human papillo-
mavirus (HPV), hepatitis C virus (HCV), Kaposi’s sarcoma-
associated herpesvirus (KSHV), and Merkel cell polyomavirus
the entire genome is maintained, as well as abortive infections
to cellular transformation and cancer development. Not surpris-
of the molecular hallmarks of cancer (10) that promote cellular
plasticity (through genomic instability, inflammation, deregula-
tion of cellular energetics, and induction of angiogenic and met-
astatic processes), proliferation (by sustaining proliferative
signaling, evading growth suppressors, and enabling replicative
lioblastoma multiforme (GBM) is an aggressive and malig-
nant tumor of glial origin with a grim prognosis (43). It ac-
immortality), and survival (avoidance of immune detection and
inhibition of apoptosis). However, many other viruses not yet
lecular events, leading to speculation that more human cancers
have a viral etiology or association than is currently appreciated.
receiving increased examination is that of HCMV and GBM.
HCMV asymptomatically infects the majority of the human
population, and virus-induced sequelae are generally observed
only under conditions of insufficient host immune function (7).
Examples include birth defects in congenitally infected neonates,
retinitis and blindness in AIDS patients, and graft rejection in
transplant patients receiving immunosuppressive therapy. How-
ever, an emerging concept hypothesizes that chronic conditions,
although perhaps not directly caused by HCMV, are likely to be
to HCMV infection include immunosenescence (40), certain car-
diovascular diseases (36), and cancer (34). Either HCMV infec-
duce all of the molecular hallmarks of cancer (13). For example,
viral infection or HCMV-encoded proteins promote genomic in-
stability (8, 30), inflammation (2), angiogenesis (4), and cell mi-
and life span (35), and inactivate cellular immune functions (25)
Received 22 August 2011 Accepted 3 November 2011
Published ahead of print 16 November 2011
Address correspondence to Robert F. Kalejta, email@example.com.
Copyright © 2012, American Society for Microbiology. All Rights Reserved.
jvi.asm.org 0022-538X/12/$12.00Journal of Virologyp. 854–864
and apoptotic pathways (3). Therefore, HCMV represents an in-
triguing candidate human cancer virus.
The major experimental method identifying HCMV in GBM
samples has been immunohistochemical (IHC) detection of the
72-kDa viral immediate-early 1 (IE1) protein. Individual studies
20 out of 21 (31), or 8 out of 49 (16). The viral phosphoprotein of
uct of the 28th gene in the unique short region of the genome
(US28) was found in 20 out of 21 GBM samples (31). In situ
hybridization (ISH) for either the HCMV IE locus (5), total viral
genomic DNA (5, 20), or an undisclosed region(s) of the viral
genome (29) has also been used to score for the presence of
HCMV, obtaining positive results in every (29 out of 29) GBM
unique long section of the genome (UL55 that encodes glycopro-
tein B [gB]) in 7 out of 9 (5) or 21 out of 34 (20) GBM specimens.
Many interpret these results as solid evidence indicating that
HCMV is present in GBMs.
detect HCMV in GBM samples. PCR and IHC for IE1 or pp65
failed to detect the presence of HCMV in 22 GBM samples (24).
Likewise, an independent study (15) that used IHC for pp65, ISH
samples. An additional investigation (28) detected IE1 by IHC in
only 9 out of 81 GBMs and by ISH in only 7 out of 81 GBMs,
cating issues have created uncertainty about the results of the
above studies that detected a strong association of HCMV with
GBM tumors. For example, the IHC images presented are very
difficult for all but a trained pathologist to decode, and thus, it
could be argued that the conclusions drawn from such experi-
ments are more subjective than objective. Furthermore, the viral
antigens are detected only when ultrasensitive IHC methods are
employed. Perhaps most troubling, the normally nuclear IE1 and
pp65 proteins detected with this technique are almost invariably
cytoplasmic when detected in GBM tissue, calling into question
the assay’s specificity. Thus, any association of HCMV with GBM
is considered with a healthy skepticism.
The previous studies discussed above often suffer from small
sample sizes, low numbers of HCMV loci analyzed, subjective
assays, and insufficient quantitative analysis. Thus, we sought to
independently determine whether HCMV was statistically more
or noncancerous brain tissues by testing for the presence of mul-
tiple viral genomic loci by PCR in GBM specimens. Our results
lead us to conclude that all regions of the HCMV genome are
present in the vast majority of GBM samples but that only a small
minority of cells in any individual sample harbors HCMV DNA.
MATERIALS AND METHODS
Sample collection and DNA extraction. This retrospective study was
conducted in agreement with the terms of a University of Wisconsin—
Madison (UW-Madison) institutional review board (IRB) protocol (M-
2009-1420). Tumor samples were obtained with prior patient consent.
The samples were deidentified and are untraceable to individual patients.
Only the date of collection of the sample and diagnosis information were
available. The samples were destroyed during the course of the analysis.
Sample preparation was performed in clean laboratories never before
used for HCMV research in order to prevent spurious contamination.
Paraffin-embedded archived astrocytoma grade 4 (GBM), meningioma,
schwannoma, oligodendroglioma, or nonneoplastic epileptic brain sam-
slides were examined. Sections containing the highest percentage of tu-
mor cells and the least amount of necrosis were chosen. The 75 GBM
samples analyzed here represent essentially the entire collection of usable
samples collected from 1994 through 2009 at UW-Madison. To purify
DNA from paraffin-embedded tissues, 10-?m sections were extracted
TABLE 1 PCR primer sequences used in this studya
Gene Forward primerReverse primer
aNucleotide sequences for primer pairs used to amplify segments of the indicated HCMV gene are shown. The expected size of the amplified product is listed in base pairs. Also,
the sequences of the primers (named “IE1”) employed in the comparative PCR (Fig. 1) and copy number (Fig. 10) studies are indicated.
HCMV Genome in Glioblastoma (GBM)
January 2012 Volume 86 Number 2jvi.asm.org 855
twice with xylene and treated with DNA lysis buffer (50 mM KCl, 10 mM
Tris-HCl [pH 8.3], 2.5 nM MgCl2, 100 ?g/ml gelatin, 0.45% IGEPAL
nase) at 55°C for 3 h and then at 95°C for 10 min. DNA was subsequently
precipitated from supernatants with sodium acetate and ethanol and fi-
nally resuspended in deionized/distilled water. For nonarchived tissue
samples, ?1 mg of frozen but not fixed tissue was minced and DNA was
manufacturer’s protocol. Positive-control samples were generated by in-
fecting primary human foreskin fibroblasts with HCMV strain TB40/E at
multiplicities of infection (MOIs) of 10?5, 10?4, 10?3, 10?2, or 10?1
PFU/cell for 18 h and preparing DNA with the Qiagen DNeasy kit. It is
suspected that little to no viral DNA replication takes place within this
time frame, so the amount of viral DNA present reflects input viral ge-
nomes. DNA was quantitated by absorbance at 260 nm.
Experimental and statistical analyses. Samples were analyzed with
PCRs and agarose-ethidium bromide gel electrophoresis in clean labora-
tories never before used for HCMV research in order to prevent spurious
contamination. For conventional PCRs, 100 ng template DNA and Taq
polymerase (New England BioLabs) along with nucleotides and buffers
provided by the polymerase manufacturer were utilized in 25-?l reaction
mixtures that were run for 35 cycles. Primer sequences are presented in
Table 1. Primer pair fidelity was confirmed by sequencing positive-
control and paraffin-embedded GBM tumor DNA reaction products. Se-
separated by electrophoresis, and bands of the expected size were quanti-
tated with densitometry using Image J software. Quantitative real-time
PCR (Q-PCR) methods have been previously described (12). Statistical
analyses were performed with Graphpad prism. Linear graphs were cre-
ated with Microsoft Excel, and heat maps were created with Heatmap
Semiquantitative PCR as an objective assay for HCMV se-
quences in GBM samples. In previous studies where PCR was
used to probe for HCMV genomes in GBM samples (5, 20, 29), a
ably, on visual examination of reaction products separated on
agarose gels. We sought a more quantitative and objective
method, and thus, we compared the efficiency with which quan-
titative real-time PCR (Q-PCR) and semiquantitative (conven-
genomes in complex samples. Templates consisted of a series of
DNA preparations from primary human fibroblasts (HFs) in-
fected with 10-fold dilutions of HCMV strain TB40/E for 18 h.
These samples also served as positive controls during the analysis
viral genomes should be complete but viral DNA replication not
yet initiated. Thus, the level of DNA detected should be directly
FIG 1 Comparison of Q-PCR and conventional PCR for the detection of an
HCMV genomic sequence. (A) Quantitative real-time PCR detection of the
HCMV UL123 (IE1) genomic locus from human fibroblasts infected at MOIs
panel A. (C) Agarose gel detection of the HCMV UL123 (IE1) genomic locus
amplified by conventional PCR using the same primers and templates as in
panel A. (D) Linear plot of Image J quantification of the gel shown in panel C.
FIG 2 HCMV genomic regions surveyed. Schematic representation of the genomic loci amplified by the PCR primers used in this study. Primer pair locations
(e.g., UL82) amplified by the indicated primer pair.
TABLE 2 P values from Wilcoxon rank sum test
P value for comparisona
replicates GBM vs epilepsy
GBM vs control
aP values for the indicated comparisons of the levels of PCR amplification products for
the indicated viral genes are displayed. P values that indicate statistical significance are
shown in boldface type.
Ranganathan et al.
jvi.asm.org Journal of Virology
related to input viral genomes. For this test comparison, primers
(Table 1) that could be used for both Q-PCR and traditional PCR
amplification of the IE1 locus were employed.
Q-PCR was performed (Fig. 1A), and threshold values were
determined and plotted (Fig. 1B) as a function of the multiplicity
of infection (MOI). Regression analysis demonstrated the linear-
ity of the assay (coefficient of determination [R2] ? 0.93). Con-
ventional PCR was also performed, and reaction products were
separated by agarose gel electrophoresis (Fig. 1C), quantitated by
Image J software, and plotted (Fig. 1D) as a function of MOI.
Regression analysis demonstrated the linearity of this assay (R2?
0.97). Thus, for this primer set, conventional PCR (yielding the
larger R2value) would have comparable (in fact, superior) pre-
dictive value for future samples than Q-PCR. Therefore, we
FIG 3 Mann-Whitney U/Wilcoxon rank sum test of paraffin-embedded samples. DNAs from tissue samples were used as templates for conventional PCR
amplification with primers detecting the indicated HCMV genomic locus. Bands on agarose gels were quantitated by Image J software, and arbitrary units are
plotted. Samples from individual classes (epilepsy, GBM, and other brain tumors [OBT]) were compared with the Mann-Whitney U/Wilcoxon rank sum test.
Statistically significant differences are indicated with asterisks as follows: ???, P value ? 0.0001; ?, P ? 0.05.
HCMV Genome in Glioblastoma (GBM)
January 2012 Volume 86 Number 2jvi.asm.org 857
judged that conventional PCR would be sufficient to obtain the
semiquantitative data that would allow for an objective evalu-
ation of the presence of HCMV DNA in a given sample. Due to
the large number of samples to be examined and primer pairs
to be employed, we chose to use conventional PCR for most of
Detection of HCMV sequences in paraffin-embedded sam-
ples. Twelve primer pairs for the examination of paraffin-
embedded samples (Table 1) were developed based on the se-
quence of the cloned TB40/E strain (GenBank accession no.
EF999921.1) that spanned 229 kb of the HCMV genome at ap-
proximately 20-kb intervals (Fig. 2). Primers within genes encod-
used whenever possible. Primer specificity was confirmed by se-
quencing PCR amplification products generated with positive-
control templates (data not shown). Paraffin-embedded tissue
sections were obtained from the Pathology archives of the Hospi-
tal & Clinics at the University of Wisconsin—Madison and con-
sisted of samples from patients diagnosed with GBM (n ? 75),
meningioma (n ? 6), schwannoma (n ? 5), oligodendroglioma
(n ? 5), or epilepsy (n ? 15). Healthy (nondiseased) brain sam-
ples were rare in the collection, so nonneoplastic brain samples
sample from the collection. DNA preparation and PCR analysis
were carried out in clean rooms not previously used for HCMV
mixtures. The total reaction products were separated by agarose
gel electrophoresis, and bands of the appropriate size were quan-
titated with Image J software (data not shown). Background val-
ues were subtracted from individual band intensities. Where in-
dicated (Table 2), technical replicates were averaged. Data and
statistical analyses are described below.
Mann-Whitney U/Wilcoxon rank sum test. The levels of
HCMV sequences in GBM, epilepsy, or other brain tumor
paraffin-embedded samples were compared using the Mann-
Whitney U test (also called the Wilcoxon rank sum test), a non-
parametric statistical examination applicable to arbitrary sample
an asterisk in Fig. 3) to indicate that GBM samples were more
likely to contain larger amounts of HCMV DNA than the com-
parison cohort and comparisons with P ? 0.0001 (denoted by
three asterisks in Fig. 3) to indicate that GBM samples were much
more likely to contain HCMV DNA than the comparison cohort.
Our analysis indicated that 8 out of 12 HCMV genomic loci
tested were statistically more likely to be present in GBM samples
2). The eight loci were UL17, UL27, UL69, UL82, UL96, UL122,
US11, and US28. Of these loci, three (UL96, US11, and US28)
were also statistically overrepresented in GBM samples compared
to other brain tumors. The HCMV UL55 locus was statistically
more likely to be present in GBM samples compared to other
brain tumors, but not compared to brain samples from patients
represented in control samples (epilepsy or other brain tumors)
compared to GBM samples. We also corrected for multiple com-
parisons by estimating false discovery rates (FDR) by calculating
Q values (1). We considered comparisons with Q ? 0.05 to indi-
cate that HCMV DNA was more likely to be found in GBM sam-
ples than in the comparison cohort and comparisons with Q ?
0.0001 to indicate that HCMV DNA was much more likely to be
of FDR analysis echoed our initial analysis (Fig. 4 and Table 2).
From these two tests, we conclude that multiple HCMV genomic
to objectively classify each sample as positive or negative for the
viral genomic region amplified by each primer pair. To obtain
objectivity for the complete data set (Fig. 5), we used a set of
while some primer sets (e.g., UL17) can be used to predict un-
known values very efficiently, others (e.g., UL55) do so less effec-
tively. At the y intercept, the value of x (the MOI of HCMV) is
Therefore, samples with band intensities greater than the
y-intercept value were considered positive, and those with band
intensities equal to or less than the y-intercept value were consid-
ered negative. Heat maps depicting either positive (red) or nega-
tive (green) results for each sample and each primer pair were
generated (Fig. 6).
fication of HCMV DNA sequences from paraffin-embedded tis-
sue was less likely in brain samples from patients with epilepsy
(Fig. 6A) and other brain tumors (Fig. 6B) than from GBM sam-
ples (Fig. 6C). By using this method of data analysis, which is
independent of the statistical one described above, we reached a
similar conclusion, namely, that multiple HCMV genomic loci,
and very likely the entire genome, are found in GBM tumors.
FIG 4 False discovery rate analysis. P values (Table 2) were converted to Q
the Q value for each primer pair is plotted for two comparisons, GBM versus
epilepsy and GBM versus OBT. Lines are drawn at P values of 0.05 and 0.001.
Values between the lines are considered to represent significant differences
between the compared cohorts. Values above both lines are considered highly
loci surveyed is shown for illustrative purposes (not to scale).
Ranganathan et al.
jvi.asm.org Journal of Virology