Epstein-Barr Virus Infection Promotes Epithelial Cell Growth by Attenuating Differentiation-Dependent Exit from the Cell Cycle

Latent infection by Epstein-Barr virus (EBV) is an early event in the development of EBV-associated carcinomas. In oral epithelial tissues, EBV establishes a lytic infection of differentiated epithelial cells to facilitate the spread of the virus to new hosts. Because of limitations in existing model systems, the effects of latent EBV infection on undifferentiated and differentiating epithelial cells are poorly understood. Here, we characterize latent infection of an hTERT-immortalized oral epithelial cell line (NOKs). We find that although EBV expresses a latency pattern similar to that seen in EBV-associated carcinomas, infection of undifferentiated NOKs results in differential expression of a small number of host genes. In differentiating NOKs, however, EBV has a more substantial effect, reducing the extent of differentiation and delaying the exit from the cell cycle. This effect may synergize with preexisting cellular abnormalities to prevent exit from the cell cycle, representing a critical step in the development of cancer.

cellular abnormalities to prevent exit from the cell cycle, representing a critical step in the development of cancer. KEYWORDS EBV, NPC, differentiation, gastric cancer, oral keratinocytes E pstein-Barr virus (EBV) is a herpesvirus that infects over 90% of the population by adulthood. Primary EBV infection usually presents as a nonspecific illness in early childhood but often manifests as infectious mononucleosis in adolescents. Thereafter, EBV establishes lifelong latent infection in B lymphocytes. Rarely, EBV latent infection results in malignancy, including Burkitt and Hodgkin lymphoma, lymphoproliferative disease, nasopharyngeal carcinoma, and gastric cancer. Much of our knowledge of the transforming effects of latent EBV genes derives from the study of latent EBV infection of B lymphocytes in vitro, which results in their growth transformation into lymphoblastoid cell lines (LCLs). Extensive investigation of LCLs has demonstrated that latent genes constitutively activate growth and survival signals essential for normal B cell development by inducing CD40-like and B cell receptor-like signaling (reviewed in reference 1).
To elucidate the mechanisms by which EBV promotes the transformation of normal epithelial cells into carcinomas, it is important to study latent EBV infection in nontransformed cells. While LCLs have been useful models for investigating EBV lymphomas, there is no analogous epithelial cell transformation model. Since EBV infection of primary epithelial cells typically results in lytic infection (2,3), the study of latent epithelial infection has been largely limited to infection of cultured carcinoma cell lines. These cell lines are of limited utility in understanding the role of latent EBV in the development of carcinomas because they cannot model critical premalignant virushost interactions. Recently, the normal oral keratinocyte (NOK) model has emerged as a nontransformed epithelial cell that is capable of supporting latent EBV infection (4).
The NOK cell line was derived by expressing human telomerase reverse transcriptase (hTERT) in cells derived from gingival tissues obtained from an oral surgery (5). These cells display a normal morphology, have a functional p53 pathway, and retain the ability to differentiate. When NOKs infected with the Akata EBV strain (NOKs-Akata) are grown in organotypic raft cultures, EBV replication is observed in the differentiated layers (6). This propensity to replicate in more-differentiated epithelial cells has been observed in clinical specimens as well as in experimentally infected primary keratinocytes (3,7). The ability of NOKs-Akata to support differentiation-dependent EBV replication has been crucial to the identification of cellular transcription factors that link these processes. For example, expression of KLF4 and BLIMP1, which promote epithelial differentiation, has been shown to induce expression of the EBV BRLF1 (Rta) and BZLF1 (Zta) immediate early genes, initiating the lytic cascade (6,8). Recently, we have shown that EBV LMP1 is directly upregulated by cellular differentiation, independently of Rta and Zta, and in fact appears to enhance Rta and Zta differentiation-dependent expression (9).
The NOKs-Akata system has also revealed effects that EBV exerts on host cells. Birdwell et al. characterized genome-wide changes in DNA methylation associated with transient EBV infection (10). Cells that had been infected displayed stable hypermethylation of CpG islands, a common characteristic in EBV-positive carcinomas. EBV infection has also been observed to alter the morphology of NOKs grown in raft cultures, resulting in disruption of the suprabasal layers and decreased expression of some makers of differentiation, such as K10 (6). In addition, in EBV-infected cells, BrdU incorporation is no longer restricted to the basal layer of cells, although it is unclear if this is due to cellular DNA replication, viral DNA replication, or both (11). Finally, EBV infection of NOKs has been associated with disruption of the basal layer, with "invasion" of some keratinocytes into the collagen matrix. This phenomenon has been linked to LEF1 upregulation (12). The EBV factors responsible for these effects, including the extent to which they are attributable to latent versus lytic EBV genes, are unknown.
In the present study, we report an analysis of global gene expression changes caused by EBV infection in NOKs-Akata. By using this unbiased approach, we expected to identify perturbations of cellular signaling pathways induced by EBV infection and to associate those changes with phenotypes previously associated with EBV infection that potentially contribute to carcinomagenesis. Because EBV gene expression is linked to differentiation, we analyzed both EBV and host cell gene expression in undifferentiated NOKs and upon differentiation by methylcellulose. Finally, using a replication-defective EBV (AkataΔRZ), we were able to distinguish the effects of EBV latent infection from those induced by viral replication.

RESULTS
EBV gene expression in NOKs-Akata resembles that seen in EBV-positive carcinomas. To examine the global changes in gene expression induced by EBV infection and differentiation, we analyzed NOKs and NOKs-Akata by transcriptome sequencing (RNA-seq). Each cell line was either grown as an undifferentiated monolayer or differentiated by suspension in methylcellulose (MC) for 24 h. Four replicates were performed for each condition from independent seedings of cells. Reads were aligned to the human genome (hg19) and to the Akata EBV strain as detailed in Materials and Methods. EBV gene expression in NOKs-Akata included the noncoding EBV-encoded RNAs (EBERs) and BamHI-A rightward transcripts (BARTs) (Fig. 1A). A variable degree of spontaneous lytic gene expression was observed in each replicate (with the strongest signals at the OriLyt, BMRF1/BMLF1, and BNLF2a/BNLF2b loci). In the absence of a differentiation stimulus, we detected very few reads mapping to the LMP1 and LMP2A/B transcripts, other than reads attributable to BNLF2a/BNLF2b, which overlap the 3= untranslated region (3= UTR) of the LMP1 transcript.
To compare EBV gene expression in NOKs-Akata to that observed in EBV-associated carcinomas, we aligned reads to the EBV genome from publicly available RNA-seq data sets of four nasopharyngeal carcinoma (NPC) (13) and 27 EBV-associated gastric cancer (EBVaGC) (14) tumors. As expected, high levels of the BART RNAs were observed. We did not reliably detect EBER expression in these samples, likely because these RNA-seq libraries were prepared from poly(A)-enriched RNA, whereas our libraries used ribodepletion. As with NOKs-Akata, there were varied amounts of expression of lytic genes. Latent membrane protein expression was highly varied; in general, LMP1 was not expressed or was expressed at very low levels in GC tumors but was present in 3 out of 4 NPC tumors ( Fig. 1B; see also Fig. S1 in the supplemental material). LMP2A/B was expressed in 16 out of 27 GC tumors and in all four NPC tumors. In addition, the BNLF2a/BNLF2b transcript was detected in 13 of 27 GC and 3 of 4 NPC tumors, which includes tumors with little or no spontaneous lytic replication. In summary, latent EBV gene expression levels varied in the tumor samples that we analyzed but conformed to a latency I/II pattern and closely resembled that observed in our NOKs-Akata model.
Differentiation affects the expression of EBV genes. We and others have previously shown that differentiation of NOKs-Akata by a variety of stimuli promotes entry into the lytic cycle (6,11), and we again observed this in our MC-treated NOKs-Akata ( Fig. 2A). Induction into the lytic cycle was more apparent after 48 h than at the 24-h time point, when RNA-seq was performed. Our RNA-seq results were remarkable for strong induction of LMP2A/B upon NOK differentiation (Fig. 2B) in 3 of 4 replicates. Examination of the unique 5= exons suggests that this was predominantly the LMP2A transcript, though lower levels of LMP2B exon 1 were present in these three replicates. The BNLF2a/BNLF2b transcript was upregulated by several orders of magnitude upon differentiation ( Fig. 2C and D). In contrast, under these conditions, we observed minimal induction of the LMP1 transcript upon differentiation of NOKs-Akata (Fig. 2D).
Few host genes are differentially regulated in NOKs infected with EBV. To determine host cell changes induced by EBV infection of undifferentiated NOKs, we compared normalized expression levels for all detected transcripts using DESeq2 ( Fig. 3A; Table S2). This identified 55 genes upregulated and 122 genes downregulated by EBV infection (Ն1.8-fold change, P Ͻ 0.05). Although these observed changes were small in magnitude and few in number, we observed the overlap of several genes previously identified as regulated by transient EBV infection in the same model ( Fig. 3B) (10). The overlap was more extensive for downregulated genes (38 of 122), whereas only 10 of 55 upregulated genes were common between the two data sets.
We also examined host gene expression induced by differentiation with MC for 24 h. Examination of several genes known to be activated during keratinocyte differentiation revealed strong induction of early markers, such as involucrin and cornifins (Table 1). However, the late differentiation markers loricrin and filaggrin were not expressed, probably a reflection of the 24-h time point chosen for our analysis. Interestingly, across all early-differentiation markers measured, expression was more strongly induced in NOKs than in NOKs-Akata. Unlike with EBV infection, differentiation resulted in differential expression of thousands of cell genes ( Fig. 4A and B). In NOKs, there were 3,946 upregulated and 3,901 downregulated genes with MC treatment. In NOKs-Akata, somewhat smaller numbers of genes were differentially regulated, with 2,752 upregu-   lated and 2,877 downregulated genes. The majority of genes affected by differentiation are common to both NOKs and NOKs-Akata (Fig. 4C). Although EBV had a limited effect on gene expression in undifferentiated NOKs-Akata, differences were much greater in the MC-differentiated state. Compared with differentiated NOKs, differentiated NOKs-Akata had 684 upregulated and 678 downregulated genes. Differentiation of both NOKs and NOKs-Akata leads to changes in the same signaling pathways, although these changes are attenuated in NOKs-Akata. To identify signaling pathways affected by MC treatment, we used gene set enrichment analysis (GSEA) with the MSigDB hallmark gene sets to characterize the regulation of genes in NOKs treated with MC and compared it with that of untreated NOKs. We observed strong downregulation of pathways associated with cell cycle progression or cell metabolism, including Myc signaling, E2F signaling, mammalian target of rapamycin complex 1 (mTORC1), and oxidative phosphorylation (Fig. 5A, blue bars). Although the MSigDB hallmark sets do not include a pathway specific to epithelial differentiation, we did observe upregulation of genes associated with transforming growth factor ␤ (TGF-␤) signaling, which has been reported to be essential for keratinocyte differentiation (15). The most strongly induced gene set was hallmark estrogen signaling early. Estrogen signaling is dependent on cell context and may contribute to both cell growth Collectively, these data suggest that the estrogen signaling early signal results from genes associated with estrogen-induced differentiation rather than growth. In raft cultures, EBV infection of NOKs results in irregular morphology and promotes invasion into the dermal equivalent (10)(11)(12). To examine how EBV may affect differentiation-induced signaling pathways, we repeated our GSEA of MC treatment with NOKs-Akata (Fig. 5A, red bars) and found a strong concordance with gene sets affected by MC in NOKs. This suggests that MC treatment affects the same signaling pathways irrespective of EBV, even though fewer genes were differentially expressed in  NOKs-Akata. To further investigate how EBV modulates differentiation, we analyzed the genes differentially expressed between NOKs and NOKs-Akata after MC treatment.
Remarkably, for almost all pathways affected by MC treatment, the effect of EBV infection was the opposite of the MC effect ( Fig. 5B; Fig. S3 and S4). The three hallmark gene sets most strongly affected by EBV were Myc targets V1, Myc targets V2, and E2F targets, all of which are strongly downregulated by differentiation but remained elevated in NOKs-Akata relative to in NOKs ( Fig. 5C to E). Because differentiation is essential for EBV to complete its replication cycle, we speculated that EBV infection would alter a subset of differentiation-induced gene changes. In an effort to identify these genes, we compared for each gene the magnitude of the MC-induced change in NOKs with that of the MC-induced change in NOKs-Akata (Fig. 6). Across all genes measured, those induced by differentiation were more highly upregulated in NOKs than in NOKs-Akata, and likewise, genes downregulated by differentiation were more highly repressed in NOKs than in NOKs-Akata. In fact, we observed a nearly perfect correlation between the MC-induced gene changes (Fig. 6, orange dashed line), with very few genes deviating substantially from the diagonal. In order to identify genes that deviate significantly from this trend, we calculated the residuals from this linear model (see the histogram in Fig. S5). Using a

EBV Attenuates Epithelial Differentiation
® likelihood ratio test, we found that 18 genes (Table S3) deviated significantly (P Ͻ 0.01). Notable among these was cyclin D3, whose upregulation in NOKs-Akata may account for global cell cycle effects of EBV upon differentiation. Akata⌬RZ is a virus mutant that is unable to undergo lytic reactivation. Because EBV lytic gene expression is activated by differentiation, we speculated that lytic gene effects, especially BGLF5-induced host shutoff (19), might be responsible for the blunting of differentiation-induced gene changes in infected cells. To determine the extent to which lytic gene expression contributes to the impaired differentiation seen in NOKs-Akata cells, we constructed a mutant virus with the BRLF1 and BZLF1 genes deleted (AkataΔRZ). These genes encode the immediate early proteins Rta and Zta, both of which are essential for EBV lytic cycle entry (20). As expected, AkataΔRZ was unable to express the lytic cascade in response to MC (Fig. 7A). Transfection of BRLF1 and BZLF1 restored the lytic cascade, including late gene expression (Fig. S6)  ingly, the BNLF2a/BNLF2b genes were expressed as latent genes and were further upregulated by MC treatment. We observed MC-induced upregulation of LMP1, as we have previously reported, but also found that LMP2A was upregulated by MC-induced differentiation independently of Rta and Zta. Unexpectedly, the AkataΔRZ virus did not express detectable BART transcripts in NOK cells. To further investigate this observation, we established multiple independent NOK clones infected with the parental Akata bacmid and AkataΔRZ. We were unable to detect BART expression in any of these cells ( Fig. 7B and C), suggesting that BART expression is defective in the original bacmid and not as a result of the BRLF1 and BZLF1 deletion. To account for this difference between our NOKs-Akata and NOKs containing a bacmid-derived virus, we used NOKs infected with the wild-type bacmid (NOKs-Akata-BAC) as a control in subsequent experiments.

Raft cultures of NOKs-Akata show impaired differentiation relative to that of NOKs even in the absence of EBV replication and BART expression.
To assess which EBV gene products account for the attenuation of differentiation, we grew NOKs and NOKs infected with wild-type Akata, Akata-BAC, or AkataΔRZ in raft cultures. In the suprabasal layers of a raft culture, NOKs display strong expression of the differentiationassociated cytokeratin K10 (Fig. 8). NOKs-Akata have substantially reduced expression of K10 throughout the suprabasal layers, expressing it most strongly in cells along the apical edge. Both NOKs-Akata-BAC and NOKs-AkataΔRZ also show a reduced expression of K10 throughout the raft, similar to that observed with NOKs-Akata.
Because one of the key steps in keratinocyte differentiation is the end of cell cycle progression, we sought to investigate the proliferative potential of rafted cells by measuring the expression of the DNA replication licensing factor MCM7. This protein is highly expressed throughout the basal layer of NOKs but is rapidly lost in suprabasal keratinocytes. However, in NOKs-Akata, NOKs-Akata-BAC, and NOKs-AkataΔRZ, MCM7 remains expressed in a subset of suprabasal cells. In addition, some cells in rafts of all three virus-infected cell lines appear below the basal layer. This can be clearly observed when rafts are stained for integrin ␤4 (Itg␤4), which anchors basal keratinocytes to the basement membrane. While staining of Itg␤4 in NOKs indicates a single, relatively smooth layer of basal cells across the surface of the collagen raft, in the virus-infected rafts, it identifies invasive growth into the collagen matrix. The invasive phenotype is observed in NOKs-Akata, NOKs-Akata-BAC, and NOKs-AkataΔRZ (Fig. 8). Thus, NOKs infected with EBV strains defective for replication and/or BART expression exhibit the EBV Attenuates Epithelial Differentiation ® same aberrant growth behavior and impaired differentiation in raft cultures as those infected with wild-type Akata.
To further assess the effects of these EBV mutations on differentiation, we assessed involucrin expression levels in three clones of NOKs, NOKs-Akata, NOKs-Akata-BAC, and two clones of NOKs-AkataΔRZ differentiated by MC for 24 h (Fig. 9). Although involucrin levels varied from clone to clone, we observed the highest levels in NOKs; levels were consistently lower in NOKs infected with any of our viruses (Akata, Akata-BAC, and AkataΔRZ).

DISCUSSION
In this paper, we report the first global analysis of gene expression changes caused by EBV infection and differentiation in a nontransformed, immortalized keratinocyte model. In undifferentiated NOKs, we observed a pattern of EBV latent gene expression that mirrors that seen in EBV-associated carcinomas. The most prominently expressed genes are the EBER and BART genes, as well as BNLF2a/BNLF2b. When NOKs were differentiated by suspension in MC for 24 h, BNLF2a and BNLF2b expression increased. In addition, LMP2A expression was induced, and a minimal amount of LMP1 expression was detectable. The dependence of these genes on cellular factors associated with   Eichelberg et al.
® differentiation likely explains their varied expression in EBV carcinomas. Specifically, LMP1 and LMP2A expression was more likely to be detected in NPC than in GC, but the levels of expression of both transcripts varied among tumor samples. LMP2A was detected in all four NPC tumors but in only 59% of GCs, while LMP1 was expressed in 3 out of 4 NPCs and mostly absent in GCs. The BNLF2a/BNLF2b transcript(s) was detected in all four replicates using NOKs and strongly induced by differentiation. We also observed the expression of BNLF2a/BNLF2b in 3 out of 4 NPCs analyzed, as well as 48% of GC tumors. This has been previously reported for GC tumors, where their expression was not correlated with lytic gene expression, suggesting that these genes may be latent in these tumors (21). Our observation of BNLF2a/BNLF2b expression in AkataΔRZ-infected cells lends additional support to this possibility. Analysis of cellular transcripts in undifferentiated NOKs suggested that latent EBV infection had only subtle effects on host gene transcription. There was overlap, particularly among downregulated genes, between genes identified in the current RNA-seq study and those detected in a previous microarray-based study wherein NOKs were transiently infected by EBV (10). This overlap is noteworthy, as these studies had significant methodological differences. In particular, the microarray study analyzed four clonal subpopulations, whereas the present study compares a population of cells with ongoing infection and against the parental clone. Since EBV-induced methylation was proposed to be the major mechanism by which transient EBV infection exerts these changes, it is interesting to note that the overlap was greatest among EBVdownregulated genes. These downregulated genes include the cadherins PCDH10, CDH11, and PCDHB6, which may contribute to the invasive phenotype, as well as DUOX1, which has been identified as playing a significant role in the regulation of differentiation-induced genes in keratinocytes (22). Among those genes upregulated by EBV in both studies are WNT5A and LEF1, which have been identified for their role in increasing cell motility and invasiveness (12).
In contrast with the minor effects of EBV infection, differentiation with MC induced much more substantial changes in gene expression. Genes associated with early keratinocyte differentiation, such as cornifins and transglutaminases, were activated, in some cases by several hundred fold. Two late-differentiation genes, filaggrin and loricrin, were not detected, possibly because treatment for 24 h was not sufficient to allow their activation. To analyze the genes affected by differentiation, we used GSEA to identify the hallmark gene sets activated or downregulated by differentiation. Gene sets that were downregulated include E2F targets, Myc targets, and a G2M checkpoint. The downregulation of these gene sets is consistent with previously established signaling events during keratinocyte differentiation (23)(24)(25)(26). Among gene sets upregulated during differentiation are those associated with an estrogen response. Estrogen signaling can be associated with either cell growth or cell differentiation. Closer inspection of the hallmark estrogen response early gene set revealed that several of the genes in this gene set that were upregulated were associated with differentiation and overlapped the Gene Ontology (GO) epithelial cell differentiation gene set. In contrast, genes in the estrogen response early gene set that are associated with cell proliferation and survival were downregulated by MC treatment. Since NOKs do not express the estrogen receptor at the mRNA level, it is likely that the hallmark estrogen response early gene set is not enriched due to estrogen signaling but because the gene set contains several genes that are expressed during keratinocyte differentiation.
Although MC treatment of both NOKs and NOKs-Akata upregulated or downregulated broadly similar genes, corresponding to enrichment of the same gene sets, there were significant differences in gene expression between NOKs and NOKs-Akata treated with MC. GSEA of the genes expressed in MC-treated NOKs and NOKs-Akata revealed a pattern that was largely the inverse of the effect of differentiation. That is, gene sets upregulated by MC treatment were lower in MC-treated NOKs-Akata than in MC-treated NOKs, and gene sets that are downregulated by differentiation are higher in NOKs-Akata compared with NOKs. The gene sets most significantly affected include two Myc target gene sets and E2F targets. These genes are strongly downregulated by differentiation as cells cease proliferation. Continued expression of these genes in NOKs-Akata likely represents a delayed exit from the cell cycle. The inverse effect of EBV on differentiation-regulated pathways indicates that the presence of EBV dampens the responsiveness to the differentiation stimulus. This is reflected in the expression levels of selected keratinocyte differentiation-associated genes, which were activated to a significantly lesser degree in NOKs-Akata. By measuring the responses of all genes to MC treatment in NOKs and comparing them with those in NOKs-Akata, we observed a broad, transcriptome-wide blunting of the differentiation effect. This suggests that EBV inhibits the differentiation process as a whole rather than disrupts only one pathway or a few signaling pathways associated with differentiation.
Our data show that the inhibition of differentiation is the most significant phenotype resulting from EBV infection of NOKs. Since differentiation of NOKs-Akata can also cause lytic reactivation of the virus, we hypothesized that expression of lytic genes may be responsible for inhibiting differentiation. To test this hypothesis, we used a recombinant virus missing the BRLF1 and BZLF1 genes, which are essential for transcription of lytic genes. The AkataΔRZ virus also did not express the BART transcript, a defect that we determined was also present in the parental bacmid virus. This may be due to the antisense insertion of the green fluorescent protein (GFP) and G418 resistance genes into a BART intron, interfering with BART transcription, unlike the insertion into the BXLF1 gene in the nonbacmid Akata. We observed the differentiation of NOKs, NOKs-Akata, NOKs-Akata-BAC, and NOKs-AkataΔRZ using organotypic raft cultures, with focus on the morphology of the rafts, expression of an early-differentiation marker (K10), and expression of a DNA replication licensing factor (MCM7). Morphologically, NOKs-Akata had less organized basal and suprabasal layers than NOKs. Some cells in the NOKs-Akata raft were observed to have grown downward into the collagen raft. This invasive behavior has been described previously (6). Invasive cell growth was readily observed when cells were stained for Itg␤4, which is strongly expressed in basal NOKs. We observed a significant degree of invasiveness across all samples of NOKs-Akata, NOKs-Akata-BAC, and NOKs-AkataΔRZ. The early differentiation marker K10 is expressed in NOKs even a single cell layer above the basal layer, and it remains expressed robustly throughout the suprabasal layers. Rafts infected with Akata, Akata-BAC, or AkataΔRZ all had substantially reduced expression of K10, and its expression was delayed until cells were more than one layer above the basal layer. In addition, some amount of MCM7 could be observed above the basal layer in all EBV-infected rafts. Taken together, these results are consistent with a delayed or inhibited differentiation in all EBV-infected rafts, regardless of the ability to express lytic genes.
Because the AkataΔRZ virus inhibited differentiation in a manner similar to that of the wild-type bacmid and nonbacmid viruses, our data suggest that EBV inhibits differentiation through a mechanism that does not require the expression of lytic genes or the BART microRNAs (miRNAs). Candidates that might mediate the inhibition of differentiation include the latent membrane proteins, the EBERs, BNLF2a/BNLF2b, or simply the presence of viral DNA. Both LMP1 and LMP2A have been reported to affect epithelial cell differentiation and growth. Expression of LMP2A in foreskin keratinocytes was reported to activate phosphatidylinositol 3-kinase (PI3K), AKT, and ␤-catenin signaling and resulted in inhibition of differentiation by suspension in methylcellulose (27), and expression of LMP2A in HaCaT cells upregulated the expression of ΔNp63␣, which is critical for the maintenance of basal keratinocytes (28,29). LMP1 has been reported to affect the growth and morphology of epithelial cells (30), although LMP1 expression is very low in our cells. The EBERs are very strongly expressed in NOKs-Akata. While their function is poorly understood, they have been described to promote cellular growth and proliferation in GC (31) and NPC (32) cell lines. While BNLF2a has been characterized to have a role in immune evasion (33,34), either gene may play a role in inhibiting keratinocyte differentiation.
By linking their replication cycles to epithelial cell differentiation, DNA tumor viruses, such as human papillomavirus (HPV) and EBV, ensure that virions are produced at epithelial surfaces, where they are best able to spread to other hosts. However, terminal differentiation presents challenges to these viruses in that the cells are exiting the cell cycle, undergoing extensive cross-linking via transglutamination of their cell cortex to cytokeratins, and ultimately becoming metabolically inert. Thus, it is not surprising that EBV has evolved strategies to attenuate differentiation to allow cellular processes to persist long enough for virions to be produced and shed. Previous work has shown that EBV induces an S-phase-like state when it begins lytic replication, allowing the replication of viral DNA (35), and that successful lytic activation depends on the activity of S-phase cyclin-dependent kinases (CDKs) (36). What is surprising is that EBV exerts these antidifferentiation effects prior to and independently of lytic cycle entry. This is consistent with prior results indicating that even transient EBV infection results in an impaired differentiation response to MC and calcium (10). We now extend these results and show that EBV exerts these effects in physiologic raft cultures and that this process does not require expression of the BARTs.
Cessation of the cell cycle is a key step in the terminal differentiation of keratinocytes (37) and can act as a failsafe to prevent uncontrolled proliferation of cells that harbor oncogenic mutations or overexpress oncogenes, such as the c-Myc gene (38). It has long been known that EBV-associated carcinomas typically exhibit defects in differentiation (39). These defects are likely a prerequisite for the maintenance of EBV in latency (40,41), as cells capable of differentiation would be negatively selected due to lytic activation of the virus (8). Our results support the notion that EBV latent genes may present additional barriers to differentiation, which may contribute to the maintenance of latent EBV infection and continued cell cycle progression. Because undifferentiated NPC is almost always EBV positive, it is likely that EBV infection is an early and rate-limiting step in its development. A direct link between EBV infection and inhibition of differentiation may represent a critical step in allowing a cell to proliferate indefinitely. Understanding how EBV-infected keratinocytes avoid differentiationassociated exit from the cell cycle is crucial to understanding EBV's role in carcinomagenesis.

MATERIALS AND METHODS
Cell lines and culture. The hTERT immortalized normal oral keratinocyte (NOKs) cell line and derivative cell lines infected with the Akata strain of EBV (NOKs-Akata) have been previously described (4,10). Additional uninfected control NOKs stably expressing enhanced GFP and G418 resistance were created by transfecting pEGFP-C1 linearized with ApaLI. The replication-defective AkataΔRZ genome was constructed in the Akata bacmid as previously described (9). All NOKs cell lines were maintained in an undifferentiated state by culturing them in keratinocyte serum-free medium (KSFM) supplemented with human epidermal growth factor and bovine pituitary extract (Life Technologies) and under selection with 50 g/ml G418.
NOKs or NOKs-Akata were differentiated following trypsinization by resuspension in 1.6% methylcellulose (MC) dissolved in KSFM and placed in 50-ml conical tubes (Falcon) for 24 h at 37°C. Cells were collected after dilution in a 1:1 volume of phosphate-buffered saline (PBS) by centrifugation (1,500 rpm for 10 min) and washed with additional PBS. Cells were grown in organotypic raft cultures as described previously (42).
RNA-seq. Cell pellets were harvested in RNA STAT-60, and RNA was isolated according to the manufacturer's instructions. RNA isolated from 10 6 cells seeded in a 60-mm culture dish served as an undifferentiated control. RNA from four independent biological replicates per condition was isolated. RNA integrity was assessed on an Agilent TapeStation 2200 using the RNA ScreenTape assay. Approximately 800 ng of total RNA was ribodepleted using the Ribo-Zero Gold reagent (Illumina). Sequencing libraries were prepared using the Illumina TruSeq stranded total RNA low-throughput (LT) kit according to the manufacturer's instructions. Library size was determined using the Agilent TapeStation 2200 D1000 assay and quantified with the Quanta PerfeCTa NGS quantitative PCR (qPCR) quantification kit. Libraries were normalized to 4 nM, pooled, denatured, and diluted to approximately 1.8 pM. A PhiX library was spiked in as an internal control. A pool of 8 samples plus PhiX was single-end sequenced using an Illumina NextSeq 500 high-output cartridge (101-bp reads).
Processing of the RNA-seq samples. RSEM (version 1.3 [43]) and Bowtie (version 1.1.1 [44]) were used to quantify gene expression using the human genome assembly hg19. DESeq2 (version 1.2 [45]) was used to analyze differences in gene expression using read counts for each gene from RSEM, excluding genes with fewer than 20 aligned reads. P values were adjusted using the Benjamini-Hochberg method. For EBV-positive samples, using Burrows-Wheeler Aligner (BWA) (46), reads were aligned to the Akata genome (GenBank accession number KC207813.1), which was modified to reflect insertion of the G418-GFP sequence into either the BXLF1 gene (47) or, in the case of the Akata bacmid, the BART locus. Alignments are summarized in Table S1 in the supplemental material. Wiggle tracks were generated from Sequence Alignment Map (SAM) alignment files and normalized for read length and millions of reads mapped (human plus EBV) to yield numbers of reads per kilobase of transcript per million mapped reads (RPKM). RNA-seq data from EBV-positive tumor samples from previously published studies (13,14) were aligned to the type I EBV reference strain (GenBank accession number NC_007605.1) using BWA. Wiggle tracks were created using the RSEM rsem-bam2wig utility.
GSEA. Gene set enrichment analysis (GSEA) was performed using the Broad Institute's GSEA implementation (48). Genes were sorted according to the signal-to-noise metric. All GSEA comparisons Eichelberg et al. were run using the hallmark gene set (49) from the Molecular Signature Database (MSigDB) (50). One replicate was not included in GSEA (NOKs-Akata [no treatment] replicate 1) due to a low number of reads uniquely aligned to the human genome. Histology. Formalin-fixed, paraffin-embedded raft sections were deparaffinized and then examined by hematoxylin and eosin and immunofluorescence (IF) staining. For IF staining, paraffin was removed in xylenes and slides were rehydrated in a series of ethanol washes followed by a PBS wash. Antigen retrieval was performed by microwaving the slides for 20 min in 10 mM sodium citrate (pH 6.0). After cooling, sections were blocked using 5% serum for 1 h and then incubated in primary antibody overnight at 4°C, followed by incubation in a 1:500 dilution of secondary antibody for 1 h and a series of PBS washes. Sections were counterstained with Hoechst 33342 dye. Secondary antibodies used were Alexa 488-conjugated donkey anti-mouse and Alexa 647-conjugated donkey anti-rabbit (Life Technologies). All images were taken with a Zeiss Axio Imager M2 microscope using the AxioVision software, version 4.8.2.
Data availability. All primary sequencing data have been submitted to the NCBI Sequence Read Archive (BioProject accession no. PRJNA555053).