1 

© The Author 2013. Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oup.com 

Statin induced changes in gene expression in EBV-transformed and native B-cells

Eugene Bolotin1, Angela Armendariz1, Kyungpil Kim1,3, Seok-Jin Heo1, Dario Boffelli1,

Kelan Tantisira2, Jerome I. Rotter4, Ronald M. Krauss1, Marisa W. Medina1,*

1Children’s Hospital Oakland Research Institute, 5700 Martin Luther King Jr. Way, Oakland

CA, 94609

2Brigham and Women's Hospital, Channing Division of Network Medicine, 181 Longwood Ave

Boston, MA, 02115

3Department of Statistics, University of California, Berkeley, CA, 94720

4Los Angeles Biomedical Research Institute at Harbor- UCLA Medical Center, Torrance, CA,

90502

*To whom correspondence should be addressed: Tel: 1-510-450-7977, Fax: 1-510-450-7909,

Email: mwmedina@chori.org

HMG Advance Access published October 30, 2013 at U

niversity of California, B

erkeley on Novem

ber 12, 2013http://hm

g.oxfordjournals.org/D

ownloaded from

2 

ABSTRACT

Human lymphoblastoid cell lines (LCLs), generated through Epstein Barr Virus (EBV)

transformation of B-lymphocytes (B-cells), are a commonly used model system for identifying

genetic influences on human diseases and on drug responses. We have previously used LCLs to

examine the cellular effects of genetic variants that modulate the efficacy of statins, the most

prescribed class of cholesterol lowering drugs used for the prevention and treatment of

cardiovascular disease. However, statin-induced gene expression differences observed in LCLs

may be influenced by their transformation, and thus differ from those observed in native B-cells.

To assess this possibility, we prepared LCLs and purified B-cells from the same donors, and

compared mRNA profiles after 24hr incubation with simvastatin (2µM) or sham buffer. Genes

involved in cholesterol metabolism were similarly regulated between the two cell types under

both the statin and sham treated conditions, and the statin-induced changes were significantly

correlated. Genes whose expression differed between the native and transformed cells were

primarily implicated in cell cycle, apoptosis, and alternative splicing. We found that ChIP-seq

signals for MYC and EBNA2 (an EBV transcriptional co-activator), were significantly enriched

in the promoters of genes up-regulated in the LCLs compared to the B-cells, and could be

involved in regulation of cell cycle and alternative splicing. Taken together, the results support

the use of LCLs for the study of statin effects on cholesterol metabolism, but suggest that drug

effects on cell cycle, apoptosis and alternative splicing may be affected by EBV transformation.

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from

3 

INTRODUCTION

Human lymphoblast cell lines (LCLs), created by transforming B-lymphocytes with Epstein Barr

Virus (EBV), have been used for functional analysis of common genetic variation by testing for

associations of genome-wide SNPs with measures of gene expression (eQTLs), chromatin state,

and mRNA transcript structure as well as the effect of copy number variation on gene expression

(1-6). In addition, LCLs have been used to identify genetic variation associated with or linked to

human diseases (7) and more recently have become a model system for pharmacogenetic studies

such as investigation of genetic effects on cellular survival in response to radiation and

chemotherapeutic drugs (8).

We have recently used human-derived LCLs to identify single nucleotide variations affecting

transcriptional response to simvastatin exposure. Simvastatin is a member of a class of drugs that

inhibit HMG-CoA reductase (HMGCR), the key regulatory enzyme in the cholesterol

biosynthesis pathway, and are widely prescribed to lower LDL-cholesterol and risk for

cardiovascular disease (9). The physiologic relevance of LCLs for use in these studies has been

demonstrated by showing that the magnitude of genetically influenced alternative splicing of the

HMGCR gene in LCLs in vitro is significantly correlated with the in vivo plasma LDL-

cholesterol response to simvastatin treatment of the individuals from whom the LCLs were

derived (10). In addition, we have successfully used the LCL model system to functionalize

haplotypes and SNPs in candidate genes (9, 11), and to identify novel genes and pathways not

previously implicated in statin effects on LDL-cholesterol (12).

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from

4 

EBV is known to infect and transform B-lymphocytes by binding to the CD21 receptor on the

cell surface (13). Upon infection it alters the cell cycle (14), impacting both the expression levels

and methylation status of hundreds of genes (15-17). Of the recent studies performed to directly

compare gene expression profiles in LCLs and B-cells (17, 18), the experiment performed by

Çalışkan et al. (15) is the most comprehensive. Examining matched B-cells and replicate LCLs

derived from six donors, the study noted that replicate LCLs generated from the same individual

clustered with each other, but clustering with progenitor B-cells was not shown. They also

reported that while transformation affected the methylation profile as well as expression levels of

genes involved in cell cycle and immune response, these changes were small in magnitude. Other

studies assessing the validity of LCLs include that of Ding et al., who reported that ~70% of

eQTLs were common to both fibroblasts and LCLs (19), suggesting that EBV transformation has

limited impact on genetic regulation of gene expression. However, in a much less well powered

study, Fairfax et al. reported that only 9.8% of eQTLs identified in LCLs were also observed in

B-cells (20). Thus, these findings emphasize the importance of understanding expression

differences in B-cells compared to LCLs, which to date, have not been examined in the context

of drug response.

In the present study, we sought to determine the effect of EBV transformation on the

transcriptional response to statin exposure by comparing statin-induced changes in the

transcriptomes of LCLs and native B-cells derived from the same individuals.

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from

5 

RESULTS

B-cells and LCLs clustering

Hierarchical clustering analysis of the genome-wide expression data that passed all quality

controls (15 statin and sham treated B-cells, and 11 statin and sham treated LCLs) demonstrated

a distinct clustering by cell type irrespective of treatment status (Fig. S1). Within each cell type

the samples further clustered by donor individual, but not by treatment status.

We then repeated this analysis after adjusting for statin response and EBV transformation using

results from the six donors for whom matched results were available (aka statin and sham treated

B-cells and LCLs). As shown in Fig. 1, we observe demonstrate significant clustering of B-cells

and LCLs from 5 of these individuals irrespective of treatment (statin or sham) status. The

mismatched cluster (individuals S_068 and S_289) had very weak bootstrap support as indicated

by numerals on branches (Fig. 1). We performed a similar analysis using genes whose

expression was significantly changed (p<0.05) by statin treatment in both B-cell and LCLs.

Although the grouping by individual was imperfect, the majority of B-cells and LCLs clustered

by individual (Fig. S2). We did not observe clustering by individual for variation of the

difference of expression levels between statin versus sham treatment (data not shown).

We then conducted a principal component analysis using the 8,000 most highly expressed genes

to assess the proportion of variation explained by EBV treatment as well as other confounders.

The first component explained 74%, of the variance and correlated highly with EBV treatment

status (P=2.2E-16). The next four components correlated with individual identities (0.005<P<

0.1) and together explained 18% of the variance. Components 6-10 were not correlated with a

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from

6 

known covariate. Finally, components 11 and 12 together explained 1.1% of the variance and

were highly correlated with statin treatment status (P<0.001) (Table S1).

Comparison of mRNA expression between B-cells and LCLs

Comparison of transcripts between B-cells and LCLs (Table S2) identified 10,304 significantly

differentially expressed genes (q<0.05). Of these, 5,491 were more highly expressed in LCLs

than in B-cells, and the remainder showed greater expression in B-cells. Genes with ≥ 2-fold

higher expression in LCLs compared to B-cells were enriched for the GO biological process term

“cell cycle” (q=1.3E-39, Fig. 2A), consistent with their transformation (Tables S3). Genes with a

smaller magnitude of up-regulation (changes between 1.32 and 2.0 fold) were enriched for the

GO term “RNA processing” (q=4.4E-14) (Tables S4).

The most down-regulated genes in LCLs compared to B-cells (≥ 2-fold lower expression) were

enriched for the GO terms “regulation of apoptosis” (q=3.3E-5) and “B-cell activation” (q=3.8E-

05) (Table S5). Genes with a more modest reduction between the two cell types (between 1.31 to

2.0 fold lower) were enriched for GO terms “mRNA metabolic processing” (q=0.03) (Table S6).

Given our interest in the use of LCLs for the study of statin pharmacogenomics and cholesterol

metabolism, we noted that, even though expression of several sterol metabolic genes such as

HMGCR, LDLR and LSS was significantly different (q<0.05) between LCLs and B-cells, overall

the category of genes involved in cholesterol and lipid metabolism was not differentially

expressed between B-cells and LCLs (q=0.17).

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from

7 

Transcripts up-regulated in LCLs are potential EBNA2 targets

We hypothesized that genes involved in EBV transformation could be direct targets of

transcriptional activators encoded within the EBV genome. Epstein-Barr Virus Nuclear Antigen

1 (EBNA1) and Epstein-Barr Virus Nuclear Antigen 2 (EBNA2) are known to be transcriptional

activators encoded by EBV (13). Re-analysis of publicly available ChIP-seq data from Raji cells

(21) (14) identified EBNA1 peaks in the promoters (-1 kb to +1 kb from transcription start site)

of 60 genes (Table S7), and EBNA2 peaks in the promoters of 2,676 genes (Table S8). 1,081

genes with EBNA2 peaks were up-regulated (q<0.05) in the LCLs compared to the B-cells, with

a highly statistically significant (P<1E-16) overlap identified between genes with EBNA2 peaks

and those up-regulated in LCLs (Table S9, Fig. 2B). GO term analysis of these genes showed an

enrichment in “cell cycle” and “RNA processing” (Tables S10, S11, S12). Among the 51 up-

regulated splicing factors only HNRNPA1 and SRFS3 (aka SRp20) have been previously shown

to interact with an essential EBV protein, SM, that modulates splicing of both EBV and host cell

pre-mRNAs (22, 23). Notably, expression levels of both SRFS3 and HNRNPA1 were higher in

LCLs compared to B-cells (1.32-fold, q=0.004 and 5.38 fold, q=3.1E-14 respectively), and these

genes had significant EBNA2 peaks in their promoters (Fig. 2C, Fig. S3, Tables S2). We also

identified a significant overlap (P=5E-05) between genes down-regulated in LCLs versus B-cells

and those with EBNA2 peaks; however there were substantially fewer genes (N=591) compared

with those that were differentially up-regulated in LCLs. Furthermore, there was no statistically

significant enrichment of GO terms within this list (Fig. S4). Overall, these results strongly

support the likelihood that EBV directly impacts host cell gene expression.

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from

8 

LCLs have reduced genome-wide methylation compared to B-cells

Next, we sought to determine if differences in gene expression between LCLs and B-cells might

be attributed to changes in DNA methylation. To test the effect of EBV transformation on

methylation profiles, we compared genome-wide methylation status of untreated LCL and B-

cells obtained from four donors. We found 2,002 regions that were significantly less methylated

in LCLs (P<0.05, paired t-test) and 756 regions that were significantly more methylated,

indicating a net effect of global de-methylation (Fig. 2D, Table S13, Table S14). Specifically,

analysis of promoter regions found evidence of reduced methylation in LCLs of 573 promoters,

and increased methylation in only 104 promoters. No enrichment of GO terms was identified in

the gene sets that demonstrated either increased or decreased methylation status between the two

cell types (Table S15).

Statin-induced changes in gene expression are similar between B-cells and LCLs

Comparison of the statin versus sham treated LCLs (N=12) identified 43 genes whose

expression was significantly changed by exposure to simvastatin (Table S16, q<0.05). This is a

much smaller number than that observed in LCLs derived from a larger study population (24), a

result to be expected based on the limitation of statistical power of this dataset. Nevertheless,

consistent with the larger study, “sterol biosynthetic process” was the top GO category

(increased expression, q=8.2E-10) identified from statin responsive genes. With the lower

threshold of P<0.01, expression of 699 genes was altered, of which 343 were up-regulated and

357 were down-regulated. GO analysis of up-regulated genes identified significant enrichment

for “sterol biosynthetic process” (q=1.5E-11), “cholesterol biosynthetic process” (q=1.32E-08)

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from

9 

and “lipid biosynthetic process” (q=0.005) (Table S17). Down-regulated GO categories included

“RNA processing” (q=1.2E-05) and “mRNA splicing” (q=0.063) (Table S18).

Again, likely due to the limited statistical power of our dataset, only 8 genes demonstrated

statistically significant changes in expression levels in the B-cells in response to statin treatment

(N=6, q<0.05): ABCG1, ZC3HAV1, ABCA1, SQLE, HMGCR, RHOB, SCD, and CYP51A1. Of

these genes only ZC3HAV1 was not previously reported to be statin responsive, while the

remainder are known to be involved in sterol metabolism and transport. At a less stringent p-

value threshold (P<0.01), 218 genes changed in response to simvastatin (Table S19), of which

123 were up-regulated and 95 were down-regulated. These included many genes known to be

sterol responsive such as LDLR and HMGCS1. GO term analysis of the 123 up-regulated genes

showed enrichment in “sterol metabolic process” (q=6.5E-08) and “cholesterol metabolic

process” (q=1.05E-05) (Table S20).

Of the genes with evidence of statin-responsiveness in both B-cells and LCLs (P<0.01), 22 were

up-regulated and one was down-regulated. (Fig. 3A, Table 1, Fig. S5). Importantly, the

magnitude of statin-induced change in expression of these genes was strongly correlated in LCLs

vs. B-cells (R2=0.723, P=4.0E-05) (Fig. 3B). Notably, these genes were primarily implicated in

cholesterol metabolism with many known to be SREBF2 targets. Among the genes that were

drug responsive in one cell type and not the other, we found enrichment of GO terms for only

“mRNA processing” and “mRNA splicing” in genes that were down-regulated by statin

treatment in LCLs but not B-cells (GO term q=4.23E-09 and q=3.0E-05 respectively).

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from

10 

Identification of transcription factors involved in statin regulation

Using Enrichr, we found that among the genes up-regulated with statin treatment in LCLs, there

was an enrichment in ChIP-seq identified binding sites for transcription factors previously

implicated in cholesterol metabolism including SREBF1 (q=9.5E-11), SREBF2 (q=1.8E-16)

EGR1 (q=2.0E-23) and HNF4A (q=3.0E-19) (25, 26). Similarly, binding sites for SREBF2

(q=4.2E-10), SREBF1 (q=2.0E-09), EGR1 (q=1.2E-06) and HNF4A (q=3.7E-06) were enriched

in statin-induced up-regulated genes in B-cells. Analysis of enriched ChIP-seq peaks in down-

regulated genes in LCLs identified 229 independent statistically enriched transcription factors

(q<0.01), of which MYC was the most significantly enriched (q=1.5E-35). In contrast, statin-

induced down-regulated genes in B-cells were not enriched for MYC binding sites, but instead

were enriched in EGR1 (q=8.1E-06), RUNX1 (q=7.2E-05) and HNF4A (7.2E-05) sites. These

results are consistent with the increased levels of MYC transcripts in LCLs versus B-cells seen in

our study (1.67 fold, q=2.83E-10), and in other studies (27, 28) (14), This raises the possibility

that MYC may play a role in statin-mediated reduction of gene expression in the LCLs but not

the B-cells. For example, HNRNPA1, a well-documented splicing suppressor (29), had strong

evidence of a MYC binding peak in its promoter and was down-regulated by statin treatment in

the LCLs, but not the B-cells (Fig. S7).

DISCUSSION

The main objective of this study was to assess the effects of EBV transformation of B-cells on

transcriptional response to drug exposure, using simvastatin as an example of a widely

prescribed class of drugs that is known to cause a robust transcriptional response through

activation of the SREBF2 pathway. In matched B-cells and LCLs obtained from the same donor

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from

11 

individuals, we documented that simvastatin exposure increased expression of many known

SREBF2 target genes involved in cholesterol metabolism in either B-cells or LCLs. Although

relatively few genes were found to be statin responsive in both B-cells and LCLs, likely due to

the limited sample size of this study, the directionality and magnitude of changes observed were

highly correlated between B-cells and LCLs. This finding suggests that EBV transformation

preserves the transcriptional response to statin treatment on either the genetic or epigenetic level.

Hence, our results support the utility of LCLs as a model system for the study of statin

pharmacogenetics.

Through hierarchical clustering analysis of the 8,000 most highly expressed transcripts identified

in our dataset we found clustering of B-cells and LCLs by individual, irrespective of statin

treatment. We further showed similar clustering using genes that were statin responsive in both

B-cells and LCLs arrays. Caliskan et al. also reported that LCLs derived from the same

progenitor B-cell population clustered by gene expression (15). These findings suggest that EBV

transformation does not disrupt the general inherent genetic or epigenetic regulation of transcript

levels in at least a subset of genes, in contrast to the report of Fairfax et al. that only 9.8% of

eQTLs identified in LCLs were also observed in B-cells (20). However, it should be noted that

our analysis was limited to a subset of transcripts and thus may not represent the transcriptome-

wide effects of EBV transformation. Furthermore, we did not observe clustering by individual

based on statin-induced differences in gene expression, either due to limited statistical power, or

to the possibility that genetic influences on the statin response are more modest than those

contributing to variation in endogenous expression levels.

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from

12 

Notably, expression of the class of genes involved in lipid metabolism was not statistically

different between the two cell types, suggesting that LCLs can be a useful model system for the

study of genetic and pharmacologic effects on cellular cholesterol homeostasis. The utility of

LCLs in this regard is further supported by the recent evidence that ≥70% of the expressed genes

in LCLs were also expressed in primary human hepatocytes and the human hepatoma cell line

HepG2 (ET, personal communication). For the genes whose expression levels differed between

B-cells and LCLs, expression of cell cycle-related genes was higher and expression of apoptosis-

related genes was lower in LCLs compared to B-cells, consistent with previous reports and the

transformed nature of the LCLs (15, 30, 31). Unexpectedly, we also identified a number of genes

involved in RNA processing and splicing whose expression was altered by EBV transformation.

These changes may be mediated at least in part by the EBV-produced transcriptional coactivator

EBNA2, as we found a significant overlap between EBNA2 ChIP-seq identified target genes and

genes up-regulated in LCLs compared to B-cells. Many of these target genes appear to be

involved in mRNA splicing, suggesting that EBV may have direct effects on the structure and

regulation of RNA generated by the host cell. Supporting this hypothesis is the observation that

EBV utilizes individual endogenous B-cell splicing factors such as SFRS3 to splice its own

transcriptome (32).

Among the genes that appear to be differentially regulated by statin treatment in the LCLs, we

found that those involved in RNA processing and splicing were down-regulated by statin

treatment in LCLs but not B-cells. While it is possible that the lack of identification of these

effects in B-cells is due to the smaller number of B-cells evaluated (7 pairs) vs. LCLs (12 pairs),

or the increased heterogeneity of the B-cell isolates vs. the LCLs, it is also possible that this may

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from

13 

represent biologically meaningful differences. For example, we found enrichment for MYC

transcription factor binding in the promoters of splicing factors and other genes down-regulated

by statin treatment in LCLs but not in B-cells. MYC is known to be up-regulated by EBV and is

essential for EBV-mediated B-cell to LCL transformation (14). Moreover, gene expression

profiling studies by others have suggested that MYC can preferentially and substantially amplify

expression of specific targets including genes involved in alternative splicing (33, 34).

Furthermore, statin treatment has been shown to down-regulate and inactivate MYC (35, 36).

These findings are consistent with our observation of MYC up-regulation by EBV

transformation, and down-regulation by statin exposure, and suggest a role for MYC in statin-

induced down-regulation of genes involved in mRNA splicing.

An example of a mechanism by which MYC may modulate statin-induced down-regulation of

genes involved in alternate splicing is provided by study of statin's effects on HNRNPA1. We

have recently shown that this RNA-binding protein, which is known to interact with the EBV

SM protein (23), modulates cholesterol response to statin by promoting alternative splicing of

HMGCR, the target of statin inhibition (Yu et al, personal communication). In the present study

we have found that HNRNPA1 is up-regulated by EBV-induced transformation, is bound by

both MYC and EBNA2, and is down-regulated by statin treatment (Fig. S6). Given our recent

observations that statin treatment causes coordinated changes in alternative splicing of multiple

genes involved in cholesterol metabolism (37), these findings highlight the complex interplay of

statin and viral interaction within LCLs.

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from

14 

This study has several caveats. First, our results are only applicable to the utility of LCLs for the

study of simvastatin-induced changes in gene expression, and cannot be generalized for effects

of other statins or drug classes. Second, while we demonstrated that gene expression ranking is

generally preserved between LCLs and B-cells, we did not specifically monitor the effect of

EBV transformation at the level of individual genes or gene variants. Third, given our

experimental design in which all LCLs were exposed to statin treatment in a single batch, while

the native B-cells were statin treated individually, it is possible that some of the expression

differences observed between cell types are due to batch effects. Finally, we did not assess the

potential influence of EBV episome load on statin-induced differences in gene expression.

However, while high (>10^6) EBV episomal load has been previously shown to be a confounder

for statin-induced changes in gene expression, others have reported that low episomal load

(thousands of copies) has little impact on gene expression in LCLs (15, 38). In this regard, we

note that the LCLs used for the present study are of a very low EBV copy number (tens to

hundreds of copies). Moreover, our principal component analyses suggests that this and other

potential confounders contribute less than 7% of the total variance of the gene expression levels.

Overall, this study supports the use of LCLs as a model system for studying differences in

cellular regulation of cholesterol metabolism in response to statin. However, since we have

identified differences between LCLs and native B-cells in expression of genes involved in cell

cycle, apoptosis, and splicing, discoveries in those pathways should be validated in alternate

model systems.

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from

15 

MATERIALS AND METHODS

B-cell isolation and EBV transformation

60ml of blood was drawn from healthy unrelated adult donors. Complete details on ethnicity,

gender and age of donors are shown in Table S21. The protocol was approved by the Children’s

Hospital Oakland Research Institute IRB and informed consent was obtained from all

participants. Peripheral blood mononuclear cells (PBMCs) were isolated from 10ml of blood

using IsoPrep (Matrix Technologies) and transformed by EBV as described previously (39, 40),

N=13. Viable cells remaining after six weeks were considered to be LCLs. PBMCs were isolated

from the remaining 50ml of blood using Lymphoprep (VWR), and B-lymphocytes were

specifically isolated from PBMCs using the B-cell Negative Isolation kit (Invitrogen) according

to the manufacturer’s protocol. Aliquots of B-cells were incubated with FITC-conjugated CD19

fluorescent antibodies and the proportion of CD19 positive cells (B-cells only) was determined

by FACS analysis (BD FACS Calibur) of 10,000 gated events. Only preparations with >90% B-

lymphocyte purity were used for further analysis.

Statin incubation and RNA isolation

B-cells and LCLs were incubated with either 2 µM activated simvastatin or a sham buffer for 24

hours in RPMI media supplemented with 500 U/ml penicillin/streptomycin, 10% FBS and 2

nmol/L GlutaMAX as previously described (10), and all cultures were maintained at 37°C with

5% CO2. The B-cells were statin or sham exposed at the time of isolation (Table S21), while

LCLs were frozen, thawed and exposed as a single batch. Simvastatin, obtained as a gift from

Merck, was activated as previously described (10). RNA was isolated from all samples and

stored at -80°C. All RNAs were converted to biotin-labeled cRNA using the Illumina TotalPrep-

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from

16 

96 RNA amplification kit (Applied Biosystems, Foster City, CA) in a single batch. cRNA was

randomized and hybridized in a single batch to Illumina HumanHT-12_V4 expression beadchips

(Illumina, San Diego, CA) at Channing Laboratory (Division of Network Medicine, Brigham

and Women's Hospital, Boston MA) according to the manufacturer’s protocols.

Gene expression analysis

Raw expression data were quantile normalized and log2 transformed using the Bioconductor

package lumi (8) and p-values for expression of all annotated probes were derived using limma

(3). Pairwise scatterplots were generated using R. Arrays with a significant number of outliers

were discarded (Figs. S7-S10). All discarded arrays were from B-cell samples, and likely reflect

impure B-cell isolates. 38 arrays from 13 donors were used to assess B-cell versus LCL

differences. Complete matched quartets of B-cell (statin and sham) and LCLs (statin and sham)

were available for 6 donors (for a complete description see Table S21). Model-based

comparisons between B-cells vs. LCLs and statin vs. sham treatment, controlled for individual,

were performed using limma (3). All analyses were performed at the probe level. For results

reported at the gene level, p-values shown reflect the most significant p-value for any probe

within that gene. Q-values (q) were generated by using the “p-adjust” function in R and refer to

FDR adjusted p-values (P). Biological process Gene Ontology (GO) analysis was conducted

using DAVID (41) using all annotated human genes as the background, and GO q-values

indicate FDR adjusted p-values. Venn diagrams were generated with Venny (42).

Hypergeometric testing implemented in R (phyper function) was used to calculate overlap p-

values in the Venn diagrams. Enrichment of ChIP-seq signal was identified using ChEA (43) and

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from

17 

the Enrichr toolkit (http://amp.pharm.mssm.edu/Enrichr/) containing a database of publicly

available ChIP-Seq results. All the analysis scripts are available upon request.

Clustering and principal component analysis

Global clustering analysis was performed using the hclust function in R on the complete

expression dataset from samples that passed quality control (15 statin and sham treated B-cells,

and 11 statin and sham treated LCLs.)

Expression arrays from the six individuals with a complete matched dataset of all four treatment

conditions (B-cell statin, B-cell sham, LCL statin and LCL sham) were filtered by subsetting on

the 8000 most highly expressed genes and adjusted for EBV transformation and statin treatment

status by linear modeling. The residuals were then clustered by the pvclust function in R, using

“correlation distance” and “average method” parameters. Confidence in the branches was

assessed by bootstrap resampling (n=1000). Principal component analysis was conducted using

the prcomp function in R, with default parameters.

Methylation assay

Genome-wide methylation status was assessed after sham only treatment in four independent

matched B-cell/LCL pairs. Strongly unmethylated islands (SUMI) or unmethylated CGs were

assayed as previously described (44, 45). Briefly, DNA was extracted by standard methods and

digested overnight with the methylation-sensitive restriction enzyme HpaII (New England

Biolabs). Indexed sequencing libraries were constructed from the HpaII digest using the standard

Illumina kit, and sequenced using 50 bp paired end reads. Table S22 shows the number of reads

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from

18 

collected for each sample that passed the quality filter. Alignments were analyzed with MetMap

to assign to each HpaII site a probability of being unmethylated p(U) and a probability of being

part of an unmethylated region p(I). MetMap annotates strongly unmethylated islands (SUMIs)

as regions in which all HpaII sites have a p(I) greater than 0.1 and where at least two HpaII

fragments within the region are represented in the MethylSeq data. SUMIs shared between all

eight samples were cross-referenced using BedTools software package (46). Statistically

significant differences in SUMI scores between LCLs and B-cells were identified using paired

Student's t-test. (Additional details can be found in the Supporting Methods).

ChIP-seq analysis

EBNA2 and EBNA1 ChIP-seq raw sequence data were reanalyzed from previous studies (14,

47). Sequences were aligned to the genome using Bowtie2 (48). Annotation, peak calling and

quality control was conducted using the HOMER package (49) with default parameters.

Although two EBNA2 datasets were available, one dataset (GSM729851) did not pass quality

control and was omitted from the analysis. Only the GSM729852 dataset was used for EBNA2

analysis and both available “input” datasets were used as background controls.

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from

19 

ACKNOWLEGEMENTS EB performed the array analysis and drafted the manuscript. AA isolated B-cells and LCLs, and

performed statin treatments and RNA-extractions. KK and EB developed and applied the PCA

clustering algorithm. SJH and DB performed the methylation assay and analysis. JIR supervised

transformation of B-cells. KT supervised the performance of expression array analysis. RMK

and MWM conceived the project, guided analysis and drafted and revised the manuscript.

We thank Elizabeth Theusch for helpful discussions. This work was supported by National Institute

of Health (1T32 HL 098057-01, U19 HL 69757, R.M.K and E.B.), (5R01 HL 1041 33-03 M.W.M), (U01

HL 065899, R01 HL 092197, R01 NR 013391 K.T.), and Dairy Research Institute (DMI#1052, R.M.K).

CONFLICT OF INTEREST STATEMENT

The authors declare that they have no conflict of interest.

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from

20 

REFERENCES

1.  Degner, J.F., Pai, A.A., Pique‐Regi, R., Veyrieras, J.B., Gaffney, D.J., Pickrell, J.K., De Leon, S., Michelini, K., Lewellen, N., Crawford, G.E. et al. (2012) DNase I sensitivity QTLs are a major determinant of human expression variation. Nature, 482, 390‐394. 2.  Dixon, A.L., Liang, L., Moffatt, M.F., Chen, W., Heath, S., Wong, K.C., Taylor, J., Burnett, E., Gut, I., Farrall, M. et al. (2007) A genome‐wide association study of global gene expression. Nat. Genet., 39, 1202‐1207. 3.  Brown, C.C., Havener, T.M., Medina, M.W., Krauss, R.M., McLeod, H.L. and Motsinger‐Reif, A.A. (2012) Multivariate methods and software for association mapping in dose‐response genome‐wide association studies. BioData Min., 5, 21. 4.  Kwan, T., Benovoy, D., Dias, C., Gurd, S., Provencher, C., Beaulieu, P., Hudson, T.J., Sladek, R. and Majewski, J. (2008) Genome‐wide analysis of transcript isoform variation in humans. Nat. Genet., 40, 225‐231. 5.  Monks, S.A., Leonardson, A., Zhu, H., Cundiff, P., Pietrusiak, P., Edwards, S., Phillips, J.W., Sachs, A. and Schadt, E.E. (2004) Genetic inheritance of gene expression in human cell lines. Am. J. Hum. Genet., 75, 1094‐1105. 6.  Stranger, B.E., Forrest, M.S., Dunning, M., Ingle, C.E., Beazley, C., Thorne, N., Redon, R., Bird, C.P., de Grassi, A., Lee, C. et al. (2007) Relative impact of nucleotide and copy number variation on gene expression phenotypes. Science, 315, 848‐853. 7.  Moffatt, M.F., Kabesch, M., Liang, L., Dixon, A.L., Strachan, D., Heath, S., Depner, M., von Berg, A., Bufe, A., Rietschel, E. et al. (2007) Genetic variants regulating ORMDL3 expression contribute to the risk of childhood asthma. Nature, 448, 470‐473. 8.  Welsh, M., Mangravite, L., Medina, M.W., Tantisira, K., Zhang, W., Huang, R.S., McLeod, H. and Dolan, M.E. (2009) Pharmacogenomic discovery using cell‐based models. Pharmacol. Rev., 61, 413‐429. 9.  Mangravite, L.M., Medina, M.W., Cui, J., Pressman, S., Smith, J.D., Rieder, M.J., Guo, X., Nickerson, D.A., Rotter, J.I. and Krauss, R.M. (2010) Combined influence of LDLR and HMGCR sequence variation on lipid‐lowering response to simvastatin. Arterioscler Thromb Vasc Biol, 30, 1485‐1492. 10.  Medina, M.W., Gao, F., Ruan, W., Rotter, J.I. and Krauss, R.M. (2008) Alternative splicing of 3‐hydroxy‐3‐methylglutaryl coenzyme A reductase is associated with plasma low‐density lipoprotein cholesterol response to simvastatin. Circulation, 118, 355‐362. 11.  Smyth, G.K., Michaud, J. and Scott, H.S. (2005) Use of within‐array replicate spots for assessing differential expression in microarray experiments. Bioinformatics, 21, 2067‐2075. 12.  Medina, M.W., Theusch, E., Naidoo, D., Bauzon, F., Stevens, K., Mangravite, L.M., Kuang, Y.L. and Krauss, R.M. (2012) RHOA is a modulator of the cholesterol‐lowering effects of statin. PLoS genetics, 8, e1003058. 13.  Young, L.S. and Rickinson, A.B. (2004) Epstein‐Barr virus: 40 years on. Nat. Rev. Cancer, 4, 757‐768. 14.  Zhao, B., Zou, J., Wang, H., Johannsen, E., Peng, C.W., Quackenbush, J., Mar, J.C., Morton, C.C., Freedman, M.L., Blacklow, S.C. et al. (2011) Epstein‐Barr virus exploits intrinsic B‐lymphocyte transcription programs to achieve immortal cell growth. Proc. Natl. Acad. Sci. U. S. A., 108, 14902‐14907. 15.  Caliskan, M., Cusanovich, D.A., Ober, C. and Gilad, Y. (2011) The effects of EBV transformation on gene expression levels and methylation profiles. Hum. Mol. Genet., 20, 1643‐1652. 16.  Min, J.L., Barrett, A., Watts, T., Pettersson, F.H., Lockstone, H.E., Lindgren, C.M., Taylor, J.M., Allen, M., Zondervan, K.T. and McCarthy, M.I. (2010) Variability of gene expression profiles in human blood and lymphoblastoid cell lines. BMC genomics, 11, 96. 

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from

21 

17.  Sun, Y.V., Turner, S.T., Smith, J.A., Hammond, P.I., Lazarus, A., Van De Rostyne, J.L., Cunningham, J.M. and Kardia, S.L. (2010) Comparison of the DNA methylation profiles of human peripheral blood cells and transformed B‐lymphocytes. Human genetics, 127, 651‐658. 18.  Carter, K.L., Cahir‐McFarland, E. and Kieff, E. (2002) Epstein‐barr virus‐induced changes in B‐lymphocyte gene expression. J. Virol., 76, 10427‐10436. 19.  Ding, J., Gudjonsson, J.E., Liang, L., Stuart, P.E., Li, Y., Chen, W., Weichenthal, M., Ellinghaus, E., Franke, A., Cookson, W. et al. (2010) Gene expression in skin and lymphoblastoid cells: Refined statistical method reveals extensive overlap in cis‐eQTL signals. Am. J. Hum. Genet., 87, 779‐789. 20.  Fairfax, B.P., Makino, S., Radhakrishnan, J., Plant, K., Leslie, S., Dilthey, A., Ellis, P., Langford, C., Vannberg, F.O. and Knight, J.C. (2012) Genetics of gene expression in primary immune cells identifies cell type‐specific master regulators and roles of HLA alleles. Nat. Genet., 44, 502‐510. 21.  Lu, F., Wikramasinghe, P., Norseen, J., Tsai, K., Wang, P., Showe, L., Davuluri, R.V. and Lieberman, P.M. (2010) Genome‐wide analysis of host‐chromosome binding sites for Epstein‐Barr Virus Nuclear Antigen 1 (EBNA1). Virol. J., 7, 262. 22.  Verma, D., Bais, S., Gaillard, M. and Swaminathan, S. Epstein‐Barr Virus SM protein utilizes cellular splicing factor SRp20 to mediate alternative splicing. J. Virol., 84, 11781‐11789. 23.  Key, S.C., Yoshizaki, T. and Pagano, J.S. (1998) The Epstein‐Barr virus (EBV) SM protein enhances pre‐mRNA processing of the EBV DNA polymerase transcript. J. Virol., 72, 8485‐8492. 24.  Lara M. Mangravite, B.E.E., Marisa W. Medina, Joshua D. Smith, Christopher D. Brown, Daniel I. Chasman, Brigham H. Mecham, Bryan Howie, Heejung Shim, Devesh Naidoo, QiPeng Feng, Mark J. Rieder, Y‐D Chen, Jerome I. Rotter, Paul M. Ridker, Jemma C. Hopewell, Sarah Parish, Jane Armitage, Rory Collins, Russell A. Wilke, Deborah A. Nickerson, Matthew Stephens, Ronald M. Krauss (2013) GATM locus is associated with reduced incidence of statin‐induced myopathy. Nature, In press. 25.  Gokey, N.G., Lopez‐Anido, C., Gillian‐Daniel, A.L. and Svaren, J. (2011) Early growth response 1 (Egr1) regulates cholesterol biosynthetic gene expression. J. Biol. Chem., 286, 29501‐29510. 26.  Yin, L., Ma, H., Ge, X., Edwards, P.A. and Zhang, Y. (2011) Hepatic hepatocyte nuclear factor 4alpha is essential for maintaining triglyceride and cholesterol homeostasis. Arterioscler. Thromb. Vasc. Biol., 31, 328‐336. 27.  Pajic, A., Polack, A., Staege, M.S., Spitkovsky, D., Baier, B., Bornkamm, G.W. and Laux, G. (2001) Elevated expression of c‐myc in lymphoblastoid cells does not support an Epstein‐Barr virus latency III‐to‐I switch. J. Gen. Virol., 82, 3051‐3055. 28.  Cherney, B.W., Bhatia, K. and Tosato, G. (1994) A role for deregulated c‐Myc expression in apoptosis of Epstein‐Barr virus‐immortalized B cells. Proc. Natl. Acad. Sci. U. S. A., 91, 12967‐12971. 29.  Smyth, G.K. (2005) Limma: linear models for microarray data. . Springer, New York. 30.  Maier, S., Staffler, G., Hartmann, A., Hock, J., Henning, K., Grabusic, K., Mailhammer, R., Hoffmann, R., Wilmanns, M., Lang, R. et al. (2006) Cellular target genes of Epstein‐Barr virus nuclear antigen 2. J. Virol., 80, 9761‐9771. 31.  Seto, E., Moosmann, A., Gromminger, S., Walz, N., Grundhoff, A. and Hammerschmidt, W. (2010) Micro RNAs of Epstein‐Barr virus promote cell cycle progression and prevent apoptosis of primary human B cells. PLoS Pathog., 6, e1001063. 32.  Cao, Z., Fan‐Minogue, H., Bellovin, D.I., Yevtodiyenko, A., Arzeno, J., Yang, Q., Gambhir, S.S. and Felsher, D.W. MYC phosphorylation, activation, and tumorigenic potential in hepatocellular carcinoma are regulated by HMG‐CoA reductase. Cancer Res., 71, 2286‐2297. 33.  Seitz, V., Butzhammer, P., Hirsch, B., Hecht, J., Gutgemann, I., Ehlers, A., Lenze, D., Oker, E., Sommerfeld, A., von der Wall, E. et al. Deep sequencing of MYC DNA‐binding sites in Burkitt lymphoma. PLoS ONE, 6, e26837. 

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from

22 

34.  Takwi, A.A., Li, Y., Becker Buscaglia, L.E., Zhang, J., Choudhury, S., Park, A.K., Liu, M., Young, K.H., Park, W.Y., Martin, R.C. et al. A statin‐regulated microRNA represses human c‐Myc expression and function. EMBO Mol. Med., 4, 896‐909. 35.  Ueta, R., Fukunaka, A. and Yamaguchi‐Iwai, Y. (2003) Pse1p mediates the nuclear import of the iron‐responsive transcription factor Aft1p in Saccharomyces cerevisiae. J. Biol. Chem., 278, 50120‐50127. 36.  Tveten, K., Ranheim, T., Berge, K.E., Leren, T.P. and Kulseth, M.A. (2006) Analysis of alternatively spliced isoforms of human LDL receptor mRNA. Clin. Chim. Acta, 373, 151‐157. 37.  Medina, M.W., Gao, F., Naidoo, D., Rudel, L.L., Temel, R.E., McDaniel, A.L., Marshall, S.M. and Krauss, R.M. (2011) Coordinately regulated alternative splicing of genes involved in cholesterol biosynthesis and uptake. PLoS One, 6, e19420. 38.  Choy, E., Yelensky, R., Bonakdar, S., Plenge, R.M., Saxena, R., De Jager, P.L., Shaw, S.Y., Wolfish, C.S., Slavik, J.M., Cotsapas, C. et al. (2008) Genetic analysis of human traits in vitro: drug response and gene expression in lymphoblastoid cell lines. PLoS genetics, 4, e1000287. 39.  Pressman, S. and Rotter, J.I. (1991) Epstein‐Barr virus transformation of cryopreserved lymphocytes: prolonged experience with technique. Am. J. Hum. Genet., 49, 467. 40.  Anderson, M.A. and Gusella, J.F. (1984) Use of cyclosporin A in establishing Epstein‐Barr virus‐transformed human lymphoblastoid cell lines. In Vitro, 20, 856‐858. 41.  Huang da, W., Sherman, B.T. and Lempicki, R.A. (2009) Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat. Protoc., 4, 44‐57. 42.  Oliveros, J.C., Vol. 2007. 43.  Lachmann, A., Xu, H., Krishnan, J., Berger, S.I., Mazloom, A.R. and Ma'ayan, A. ChEA: transcription factor regulation inferred from integrating genome‐wide ChIP‐X experiments. Bioinformatics, 26, 2438‐2444. 44.  Singer, M., Boffelli, D., Dhahbi, J., Schonhuth, A., Schroth, G.P., Martin, D.I. and Pachter, L. MetMap enables genome‐scale Methyltyping for determining methylation states in populations. PLoS Comput. Biol., 6, e1000888. 45.  Vera, J. and Torres, N.V. (2004) MetMAP: an integrated Matlab package for analysis and optimization of metabolic systems. In Silico Biol., 4, 97‐108. 46.  Quinlan, A.R. and Hall, I.M. (2010) BEDTools: a flexible suite of utilities for comparing genomic features. Bioinformatics, 26, 841‐842. 47.  Dresang, L.R., Vereide, D.T. and Sugden, B. (2009) Identifying sites bound by Epstein‐Barr virus nuclear antigen 1 (EBNA1) in the human genome: defining a position‐weighted matrix to predict sites bound by EBNA1 in viral genomes. J. Virol., 83, 2930‐2940. 48.  Langmead, B. and Salzberg, S.L. (2012) Fast gapped‐read alignment with Bowtie 2. Nat. Methods, 9, 357‐359. 49.  Heinz, S., Benner, C., Spann, N., Bertolino, E., Lin, Y.C., Laslo, P., Cheng, J.X., Murre, C., Singh, H. and Glass, C.K. (2010) Simple combinations of lineage‐determining transcription factors prime cis‐regulatory elements required for macrophage and B cell identities. Mol. Cell, 38, 576‐589.  

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from

23 

LEGENDS TO FIGURES

Figure 1. Clustering of B-cells and LCLs by individual

Expression array values for B-cells and LCLs were quantile normalized, adjusted for treatment

status and clustered using pvclust in R. Individual IDs designated with S# are shown to co-

cluster. Numbers on the branches are boot strap support (N=1000).

Figure 2. Global gene expression and methylation in B-cells versus LCLs

(A) Genome-wide gene expression was quantified by 38 arrays in statin and sham treated B-

cells and LCL, (7 B-cell and 12 LCL individuals), and the resulting data was analyzed using

limma. The genes changed in response to statin treatment (q< 0.05) were classified into “high

response” (>2 fold) and “low response” (< 2 fold) groups, and subject to GO term analysis using

DAVID. (B) Overlap of genes up-regulated in LCLs and EBNA2 ChIP-seq targets. EBNA2

ChIP-seq data was reanalyzed using HOMER, and GO term enrichment was conducted using

DAVID. Statistical analysis demonstrating significant (P < 1E-16) overlap of the two gene sets

was generated using the phyper function in R. (C) ChIP-seq and expression profiling of

SFRS3. The peak diagram depicts the statistically significant (P=9.77E-08) EBNA2 ChIP-seq

peak found in the SFRS3 promoter. The H3K27Ac ChIP-seq ENCODE track, which marks

promoter regions, is shown for reference. The boxplot depicts normalized values of SFRS3

expression in B-cells (N=7) versus LCLs (N=12). (D) Violin plots of distribution of SUMI

sites in B-cells and transformed LCLs. Genome-wide methylation in B-cells and LCLs derived

from the same donor (N=4 matched pairs) was quantified by MetMap. Y-axis: 1 is completely

methylated and 0 is completely un-methylated. X-axis: paired samples.

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from

24 

Figure 3. Genes affected by statin treatment in B-cells and LCLs.

(A) Overlap analysis of statin-responsive genes in B-cells and LCLs. Genes changing in

response (P<0.01) to statin treatment in B-cells (N=7) and LCLs (N=12) (Table S1) were

identified using limma. GO term analysis was performed using DAVID. (B) Correlation of

statin-induced changes in gene expression paired B-cells and LCLs. The average magnitude

of change in genes identified to be statin responsive (P<0.01) in both B-cells and LCLs (N=23

genes) was plotted and linear regression used to assess their relationship.

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from

25 

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from

Figure 2. Bolotin et al. 2013

LCLs – B-cell

fold change

>2 Cell Cycle

Enriched

GO terms

mRNA Splicing 0 to 2

mRNA Metabolic

Process (splicing) 0 to -2

< -2 Apoptosis

B-cell activation

A

5491 genes

5046 genes

B

1081 4410

Cell Cycle

+

RNA splicing

RNA processing Genes

Up-regulated

in LCLs

Genes with

EBNA2

ChIP-seq peaks

1595

C

LCL B-cells

Expre

ssio

n A

U

EBNA2 ChIP-seq

H3K27Ac ChIP-seq

D

B-cells LCL

Deg

ree o

f m

eth

yla

tio

n

1 2 3 4 1 2 4 3

0.0

0.2

0.4

0.6

0.8

1.0

SRSF3

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from

27 

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from

28 

at University of C

alifornia, Berkeley on N

ovember 12, 2013

http://hmg.oxfordjournals.org/

Dow

nloaded from