A meta-analysis of array-CGH studies implicates antiviral immunity pathways in the development of hepatocellular carcinoma.

Abstract

The development and progression of hepatocellular carcinoma (HCC) is significantly correlated to the accumulation of genomic alterations. Array-based comparative genomic hybridization (array CGH) has been applied to a wide range of tumors including HCCs for the genome-wide high resolution screening of DNA copy number changes. However, the relevant chromosomal variations that play a central role in the development of HCC still are not fully elucidated.
In present study, in order to further characterize the copy number alterations (CNAs) important to HCC development, we conducted a meta-analysis of four published independent array-CGH datasets including total 159 samples.
Eighty five significant gains (frequency ≥ 25%) were mostly mapped to five broad chromosomal regions including 1q, 6p, 8q, 17q and 20p, as well as two narrow regions 5p15.33 and 9q34.2-34.3. Eighty eight significant losses (frequency ≥ 25%) were most frequently present in 4q, 6q, 8p, 9p, 13q, 14q, 16q, and 17p. Significant correlations existed between chromosomal aberrations either located on the same chromosome or the different chromosomes. HCCs with different etiologies largely exhibited surprisingly similar profiles of chromosomal aberrations with only a few exceptions. Furthermore, the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis indicated that the genes affected by these chromosomal aberrations were significantly enriched in 31 canonical pathways with the highest enrichment observed for antiviral immunity pathways.
Taken together, our findings provide novel and important clues for the implications of antiviral immunity-related gene pathways in the pathogenesis and progression of HCC.

Full text

Page 1

A Meta-Analysis of Array-CGH Studies Implicates
Antiviral Immunity Pathways in the Development of
Hepatocellular Carcinoma
Xu Guo1., Yanna Ba2., Xi Ma1, Jiaze An3, Yukui Shang1, Qichao Huang1, Hushan Yang4, Zhinan Chen1*,
Jinliang Xing1*
1State Key Laboratory of Cancer Biology, Department of Cell Biology, Cell Engineering Research Center, The Fourth Military Medical University, Xi’an, People’s Republic of
China, 2Department of Clinical Immunology, Xijing Hospital, The Fourth Military Medical University, Xi’an, People’s Republic of China, 3Department of Hepatobiliary
Surgery, Xijing Hospital, The Fourth Military Medical University, Xi’an, People’s Republic of China, 4Division of Population Science, Department of Medical Oncology,
Kimmel Cancer Center, Thomas Jefferson University, Philadelphia, Pennsylvania, United States of America
Abstract
Background: The development and progression of hepatocellular carcinoma (HCC) is significantly correlated to the
accumulation of genomic alterations. Array-based comparative genomic hybridization (array CGH) has been applied to a
wide range of tumors including HCCs for the genome-wide high resolution screening of DNA copy number changes.
However, the relevant chromosomal variations that play a central role in the development of HCC still are not fully
elucidated.
Methods: In present study, in order to further characterize the copy number alterations (CNAs) important to HCC
development, we conducted a meta-analysis of four published independent array-CGH datasets including total 159
samples.
Results: Eighty five significant gains (frequency $25%) were mostly mapped to five broad chromosomal regions including
1q, 6p, 8q, 17q and 20p, as well as two narrow regions 5p15.33 and 9q34.2-34.3. Eighty eight significant losses (frequency
$25%) were most frequently present in 4q, 6q, 8p, 9p, 13q, 14q, 16q, and 17p. Significant correlations existed between
chromosomal aberrations either located on the same chromosome or the different chromosomes. HCCs with different
etiologies largely exhibited surprisingly similar profiles of chromosomal aberrations with only a few exceptions.
Furthermore, the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis indicated that the genes affected by
these chromosomal aberrations were significantly enriched in 31 canonical pathways with the highest enrichment observed
for antiviral immunity pathways.
Conclusions: Taken together, our findings provide novel and important clues for the implications of antiviral immunity-
related gene pathways in the pathogenesis and progression of HCC.
Citation: Guo X, Ba Y, Ma X, An J, Shang Y, et al. (2011) A Meta-Analysis of Array-CGH Studies Implicates Antiviral Immunity Pathways in the Development of
Hepatocellular Carcinoma. PLoS ONE 6(12): e28404. doi:10.1371/journal.pone.0028404
Editor: Shree Ram Singh, National Cancer Institute, United States of America
Received October 19, 2011; Accepted November 7, 2011; Published December, 2011
Copyright: ? 2011 Guo et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted
use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: This work was supported by grant 30872927 from the National Natural Science Foundation of China and grant 2009CB521704 from the National Basic
Research Program. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing Interests: The authors have declared that no competing interests exist.
* E-mail: xingjl@fmmu.edu.cn (JX); znchen@fmmu.edu.cn (ZC)
. These authors contributed equally to this work.
Introduction
The development and progression of hepatocellular carcinoma
(HCC) is significantly correlated to the accumulation of genomic
alterations [1]. Therefore, it is important to have a clear landscape
of the genomic aberrations that occur during the multistep process
of hepatocarcinogenesis. Previous studies have used high-resolution
molecular karyotyping analyses to provide a comprehensive catalog
of structural aberrations of the whole chromosomes in HCC [2].
However, this method is highly specialized and time-consuming. As
a consequence, only a very limited number of HCC cases have been
evaluated in these studies. Moreover, the modest resolution of the
karyotyping analysis made it difficult to fully define the overall
genomic profiles of HCC in a more accurate manner. Comparative
genomic hybridization (CGH) has been developedin recent years to
monitor the DNA copy number changes at a global level [3].
However, traditional CGH techniques still have the limitation of
modest resolution (approximately 2 Mb for amplifications and 10–
20 Mb for deletions) and thus could not detect changes in smaller
chromosomal regions [4]. In comparison, array-based CGH (array
CGH) is a newly developed technology that allows for high-
throughput and high-resolution (at 1 Mb) screening of genome-
wide DNA copy number changes (either amplifications or deletions)
at the gene level [5]. Array CGH combines fluorescence techniques
PLoS ONE | www.plosone.org1December 2011 | Volume 6 | Issue 12 | e28404
12

Page 2

with the microarray platform that allows for the comparison of
DNA content in two differentially labeled genomes: a test genome
(patient)and a referencegenome(control).Themicroarray platform
also allows for the simultaneous scanning of thousands of individual
DNA sequences from the whole genome, and provides high-
resolution data on the locations of identified aberrations in a single
experiment. To date, array-CGH has been applied to a wide range
of solid tumors, including liver, breast, gastric, kidney and bladder
cancers [6,7,8,9,10]. Recently, another technology platform based
on single nucleotide polymorphism (SNP) array has been developed
to determine the copy number abnormalities of genomic DNA at
sub-kilobase resolution [11,12]. Except for an advantage of high
resolution, this platform also has a limitation of high signal-to-noise
ratio which is hard to improve[13].
Many investigators have made varying attempts to search for
genes implicated in hepatocarcinogenesis. Screening for chromo-
somal regions with frequent gains and losses is one of the first steps
toward the identification of genes. Using the traditional and array-
CGH, frequent DNA copy number gains at chromosomes 1q, 8q
and 20q, and frequent DNA copy number losses at 1p, 4q, 8p,
13q, 16q and 17p have been identified in HCC samples
[6,14,15,16,17,18,19]. Some of these regions contain known
candidate oncogenes or tumor suppressor genes, such as
ZNF217 (20q13) [20], TP53 (17p13), RB1 (13q14) [21]and cyclin
D1 (11q13) [22]. However, it is believed that the currently
identified genes represented only a small percentage of causal
elements in hepatocarcinogenesis and the vast majority of genes
with chromosomal aberrations that may play a central role in
HCC development are still unknown.
Meta-analysis is a systematic and quantitative synthesis of prior
evidence [23]. It offers the opportunity to critically evaluate and
statistically combine the results of comparable studies or trials in
order to achieve more robust and reliable results as well as identify
novel findings that are not apparent in individual studies. In
previous reports, a meta-analysis of CGH data comprising of 785
HCCs has been carried out and identified significant correlations of
chromosomal deletions on 4q, 13q, and 16q with hepatitis B virus
(HBV) etiology [24]. Recently, using the array-CGH technology,
several different studies have generated a wealth of data on more
than 100 analyzed HCC samples that await a more comprehensive
interpretation [16,25,26,27]. The aim of this study was to identify
potential genes and pathways important to HCC by utilizing the
available data from published array CGH studies of human HCC.
Materials and Methods
Data collection of array CGH studies in HCC
Datasets for HCC array CGH studies were identified from
public resources including the supplementary files of published
papers, NCBI Gene Expression Omnibus (GEO, http://www.
ncbi.nlm.nih.gov/geo), ArrayExpress (http://www.ebi.ac.uk) da-
tabase using hepatocellular carcinoma and array-based compar-
ative genomic hybridization as keywords. Datasets from studies
using HCC cell lines were excluded. We identified four datasets
with complete original data that are publicly available, including
two from the supplementary files of published HCC array CGH
studies and two from the GEO database (GSE8351 and
GSE22635) [16,25,26,27]. The detailed information was listed in
Table 1. Three of the four studies used BAC clone as the
hybridization probe while one used synthetic oligonucleotides as
the hybridization probe. A total of 159 HCC tissue samples in
these four datasets were collected, including 54 samples with HBV
infection, 57 with hepatitis C virus (HCV) infection, 6 with the
infections of both HBV and HCV, 3 with positive hepatitis B virus
X protein (HBx), and 39 samples without viral infection.
Data pre-processing for the integration across different
platforms
Because the four array CGH datasets in this study were
generated using different types of technical platforms that contain
different numbers of probes at varied spacing and resolution, they
cannot be directly compared and combined. To transform the
datasets from different platforms into a common format for the
purpose of meta-analysis, we pre-processed the original dataset of
each study based on a procedure previously described with minor
modification [24,28]. The detailed procedure used in this study is
described below:
First step: reconciliation of genome mapping data
generatedfromBACclones
probes.
In the original datasets, chromosomal positions of
BAC and oligo probes were assigned based on the different
versions of human genome assembly, such as hg15, hg17 and
hg18. Therefore, we used the annotation database of UCSC
human genome (hg19/GRCh37) to re-assign the start and end
chromosomal positions for all BAC and oligo probes from the four
datasets. The unmappable probes were excluded from further
analysis.
Second step: assignment of chromosomal positioning
anchors.
Because the copy number alterations (CNAs) of
chromosomal segments were detected on different scales by
different probes in the four datasets, it was difficult to directly
compare and integrate these datasets. To resolve this issue, we
assigned a set of chromosomal positioning anchors that were
composed of the start and end chromosomal positions of all the
probes used in the four datasets. The log2-transformed DNA copy
number ratio for each anchor was then determined for all the
samples from different datasets based on the following principles.
andoligonucleotide
Table 1. Information of 4 collected public datasets (n=159).
ID ReferencesPlatform
Etiology, N
AllHBV HCVHBV/HCV HBxnon-viral
1 Mohini A.Patil et al.(2005) BAC44343——7
2 Yasuyo Chochi et al.(2009)BAC 426*34* 2*——
3 Christof Schlaeger et al.(2008) BAC63 11144331
4Kazuya Taniguchi et al.(2010)Oligo1036——1
Total159 545763 39
*Information of virus infection for individual samples is unavailable. HBV, hepatitis B virus; HCV, hepatitis C virus; HBx, hepatitis B virus X protein.
doi:10.1371/journal.pone.0028404.t001
Array CGH Meta-Analysis in HCC
PLoS ONE | www.plosone.org2 December 2011 | Volume 6 | Issue 12 | e28404

Page 3

For any dataset, when anchors lied in the probes of the dataset, the
log2-transformed ratio for probe was assigned to corresponding
anchors. When an anchor lied outside of any detected probes of
the dataset, a missing value was created and set to the anchor.
Using this approach, we were able to delineate the same
chromosomal positions to the different probes used in the
different datasets of the four studies.
Third step: scaling and segmentation across multiple
platforms.
To integrate data across multiple sources for meta-
analysis, multi-platform segmentation with proper scaling was
performed by using a multi-platform circular binary segmentation
(MPCBS) algorithm recently reported by Zhang et al [28], which
was implemented into an R package which can be accessed on R-
Forge undera projectname MPCBS(http://r-forge.r-project.org/).
This algorithm, which is based on a simple multi-platform change-
point model, relies on a weighted sum of t-statistics to scan for copy
number changes and does not require a pre-standardization of
different data sources. To minimize the computational complexity
and maximize the generality, chromosomes were segmented
individually and sex chromosomes were not included.
Fourth step: determination of gain and loss events.
segmentation, theCNAs of
determined using a nonhierarchical k-means clustering (k=3)
algorithm [29]. All log2-transformed ratio values of the segments
generated in the third step were clustered and classified into three
subgroups bytwothresholdvalues(0.141and20.136),representing
copy number gain (.0.141), copy number loss (,20.136), and no
copy number change (between 20.136 and 0.141).
Fifthstep: evaluation
mance.
Pearson Correlation Coefficient (PCC) analysis was used
to evaluate the performance of pre-processing. We first calculated the
PCC value of log2-transformed ratio between any two samples from
the same datasets. Then we estimated the correlation among samples
of different platforms before and after pre-processing. The PCC
values of any two samples should not show the dependencies on the
platform after high-quality preprocessing.
After
were chromosomalsegments
of pre-processing perfor-
Profiling of chromosomal aberrations and pathway
analysis
Based on the definition of gain and loss events, we conducted
genome-wide profiling of CNAs for all the HCC samples included
in the four datasets and calculated the frequencies of chromosomal
aberrations for each sample. We then, using a Chi-square test,
compared the frequencies of copy number gains and losses
separately between HBV-HCC and HCV-HCC, or between
HCC with and without viral infection. In further analyses, we only
focused on those highly prevalent events of copy number gains and
losses (frequency .25%). We first evaluated the correlation
between any two CNAs by calculating the PCC value. Then,
the genes in these segments with frequent gains/losses were
mapped into the pathways derived from the Kyoto Encyclopedia
of Genes and Genomes (KEGG) database [30]. The pathways that
were significantly affected by the identified CNAs were determined
by Fisher’ exact test.
Results
Profile of chromosomal alterations in all HCC samples
We profiled the CNAs (gains or losses) of chromosomal
segments in human HCC through a meta-analysis of four
available independent array CGH datasets including total 159
samples. To integrate all datasets from different platforms, the raw
data were pre-processed and the pre-processing performance was
evaluated using the PCC analysis. Our results indicated that the
PCC values of individual samples within a platform were
significantly higher than those between two different platforms
before data pre-processing. After pre-processing, the correlation
obtained for samples between different platforms after pre-
processing was significantly improved with an increase of average
correlation coefficient from 0.198 to 0.349 (Fig. 1). These data
showed that the four independent datasets have been successfully
transformed into a comparable form. Next, chromosomal gain and
Figure 1. Comparison of correlations among log2-transformed ratios of 159 samples from four independent datasets before (a)
and after preprocessing (b). In triangle (a) and (b), color points dotted in rectangles represent PCC values of log2-transformed ratio between any
two samples from different datasets, while color points dotted in small triangles represent those from same dataset. The brightness of color blue is
directly proportional to the value (0 – 1) of Pearson correlation coefficient.
doi:10.1371/journal.pone.0028404.g001
Array CGH Meta-Analysis in HCC
PLoS ONE | www.plosone.org3December 2011 | Volume 6 | Issue 12 | e28404

Page 4

loss events were determined using log2-transformed ratio thresh-
olds of 0.141 and -0.136 derived from the k-means clustering
analysis and the frequencies of chromosomal aberrations were
calculated for all HCC samples. Our results indicated that 85
chromosomal segments exhibited a frequent copy number gain
($25%) and 88 exhibited a frequent copy number loss ($25%).
Detailed data was not presented here because of the space
limitation. They are available when required. As shown in Fig. 2,
the significant gains were mostly mapped to several broad
chromosomal regions including 1q, 6p, 8q, 17q and 20q, as well
as four narrow regions including 5p15.33- 5p14.2, 9q34.2-34.3,
20p13 and 20p11.21. Significant losses were frequently observed
in 4q, 6q, 8p, 9p, 13q, 14q, 16q, and 17p. Among these regions,
chromosome 1q was most significantly affected by copy number
gain and chromosome 8p was most significantly affected by copy
number loss. In particular, gains in 1q32.1, 1q24.1, and 1q21.3-
23.3 were detected in .70% of all HCC samples, and losses in
8p23.3-21.3 were detected in .55% of the samples.
Correlations between significant chromosomal
aberrations
Spearman’s rank correlation coefficient was calculated to assess
the correlations between different chromosomal copy number
gains and losses described above at the significant level of
p,0.001. Our results indicated that significant correlations existed
between specific chromosomal aberrations either located on the
same or different chromosomes and the significances from the
same chromosome were usually higher than those from different
chromosomes. As shown in Fig. 3(a), significant finding from the
same chromosome were mainly identified for chromosomes 1q,
4q, 5p, 6p, 6q, 8p, 8q, 9p, 13q, 14q, 16q, 17p, 17q, 20p, 20q and
between 6p and 6q, 8p and 8q, 17p and 17q, 20p and 20q. As
indicated in Fig. 3(b), significant correlations mainly existed
between 1q gains and 5p gains, and 6p gains or 4q losses, with a
positive correlation coefficient of at least 0.25, 20q gains and 9p
losses or 14q losses with a negative correlation coefficient of over
20.3. In addition, losses of 4q, 9p, 13q, 16q and 17p frequently
co-occurred with a correlation coefficient ranging from 0.26 to
0.46. Significant correlations were also observed between 6p gains
and 8p losses, 8p losses and 9p losses, and 14q losses and 16q
losses.
Canonical pathways significantly linked with
chromosomal aberrations
To explore the potential effects of chromosomal aberrations
implicated in the molecular mechanism of HCC development, we
further analyzed the functional KEGG pathways enriched with
genes located on the chromosomal segments with significant copy
number alterations. Our results indicated that the genes affected
by chromosomal aberrations were significantly enriched in 31
canonical pathways. The list of pathways together with the related
gene information was summarized in Table 2. According to the
Figure 2. Profile of chromosomal alterations in all HCC samples (n=159). (a), heatmap of CNAs across all chromosomes. (b), the
frequencies of CNAs across all chromosomes. Copy number gain and loss events of chromosomal segments were determined using log2-transformed
ratio thresholds of 0.141 and 20.136 derived from the k-means clustering analysis. The color brightness is directly proportional to the frequency of
CNA. Color red represents copy number gain and color blue represents copy number loss. CNA: copy number alteration.
doi:10.1371/journal.pone.0028404.g002
Array CGH Meta-Analysis in HCC
PLoS ONE | www.plosone.org4 December 2011 | Volume 6 | Issue 12 | e28404

Page 5

biological function, these pathways were divided into three major
categories: 18 immune-related pathways, 3 cancer-related path-
ways and 10 metabolism-related pathways. Among these path-
ways, antiviral immunity pathways were most significantly
affected, such as the antigen processing and presentation pathway
(hsa04612), RIG I like receptor signaling pathway (hsa04622),
natural killer cell mediated cytotoxicity pathway (hsa04650),
cytosolic DNA sensing pathway (hsa04623), toll like receptor
signaling pathway (hsa04620), and cytokine and cytokine receptor
interaction pathway (hsa04060). Moreover, we found many
pathways were interrelated and driven by the presence of same
or similar sets of genes. For example, seven immune-related
pathways (hsa05320, hsa04612, hsa04622, hsa04650, hsa04623,
hsa04620 and hsa04140) were driven by signals in the same core
set of the interferon-alpha family genes (IFNA1, IFNA10, IFNA13,
IFNA14, IFNA16, IFNA17, IFNA2, IFNA21, IFNA4, IFNA5, IFNA6,
IFNA7, IFNA8) that are contiguously located within a 0.3 Mb
region on chromosome 9p21.3. These genes play a vital role in the
non-specific antiviral immunity that occurs at the early phase of
viral infections. We also found that 10 immune-related pathways
harbored another core set of genes, the human leukocyte antigen
gene family [27]. In addition, six metabolism-related pathway
(hsa00982, hsa00140, hsa00053, hsa00983, hsa00980, hsa00040)
had the same core set of genes, the uridine diphosphate
glycosyltransferase2 family(UGT2A1,
UGT2B7,UGT2B10,UGT2B11,
UGT2B28), that are contiguously located within a 2 Mb region
on chromosome 4q13.2. Only a few cancer-related pathways were
found to be significantly affected by the chromosomal aberrations
identified in this study with a common gene set. However, several
oncogenes and tumor suppressor genes are mapped to these
pathways, such as TP53, RB1, and MYC, which have been shown
to play an important role in hepatocarcinogenesis [21,31].
UGT2A3,
UGT2B15,
UGT2B4,
UGT2B17,
Etiology-related chromosomal aberrations
We investigated the profiles of etiology-related chromosomal
aberrations in HCCs. As indicated in Fig. 4, HCC samples with
different etiologies exhibited very similar profiles of chromosomal
aberrations with only a few exceptions. When we compared the
profiles of 48 HBV-related and 23 HCV-related HCCs, 14 copy
number gains and 59 losses had significantly different frequencies.
Almost all of these segmental gains and losses were harbored in
HBV-related HCCs, except for gains of 12p13.1-12p12.3 and
losses of 3q26.1-3q26.2 and 3q26.2-3q26.33 that appeared in
HCV-related HCCs with significantly higher frequencies. The
significant copy number gains and losses in HBV-related HCCs
were most commonly located on chromosomal region 10p, and
14q and 4q, respectively. We also performed the comparison of
chromosomal aberration profiles between virus-related HCCs and
non-virus-related HCCs, and identified 27 gains and 17 losses
showing a significant difference in frequencies. All significant losses
were present in virus-related HCCs and most were located on
chromosomal region 16q. In comparison, significant gains were
mostly located on chromosomal regions 2p and 4p in virus-related
HCCs, whereas chromosomal regions 8p and 10p in non-virus-
related HCCs.
Discussion
Meta-analysis of array CGH is valued as an integrated
approach to simultaneously analyze information from different
studies for the detection of chromosome aberrations with
enhanced resolution and accuracy. A recent study has successfully
performed a meta-analysis of array CGH data derived from a
series of primary cancers to cluster tumor type by CNAs [24]. In
this study, we, for the first time, conducted a cross-platform meta-
analysis of HCC array CGH data obtained from different studies.
Figure 3. Correlations between significant chromosomal aberrations either located on the same (a) or different (b) chromosomes in
all HCCs. Chromosomal segments with significant gain were highlighted in red and those with significant loss in blue. Spearman’s rank correlation
coefficient was calculated to assess the correlations between different chromosomal copy number gains and losses at the significant level of
p,0.001.
doi:10.1371/journal.pone.0028404.g003
Array CGH Meta-Analysis in HCC
PLoS ONE | www.plosone.org5December 2011 | Volume 6 | Issue 12 | e28404