This study used modern molecular methods to examine a large panel human and viral RNAs in gastric cancer. To our knowledge, this is the largest panel of viral gene products to be examined in concert with human RNAs in archival, paraffin embedded tissues. The EBV-infected subtype of gastric cancer is dramatically evident in the corresponding heat map created by unsupervised clustering, and EBV infection was confirmed by high EBV DNA viral loads in these tissues. Expression of selected viral and human genes in the cancers confirmed several known virus- and cancer-related effects and also revealed novel findings that shed light on pathogenesis and possible disease management strategies.
Surprisingly, the infected gastric cancers overexpressed all 18 of the latent and lytic EBV genes that were tested. We discovered high levels of BRLF1 RNA (encoding the immediate early viral protein triggering lytic replication in concert with BZLF1) and moderately high levels of BXLF1 (the viral thymidine kinase that converts penicyclovir to a toxic form, suggesting a mechanism for therapy)
. BLLF1 (encoding the late viral envelope protein gp350/220) was expressed at moderate levels that were nevertheless significantly higher than in non-malignant mucosa, suggesting that EBV lytic infection is not abortive but rather is capable of producing the late viral envelope protein gp350/220. Among the latent genes, EBNA1 from the Q promoter, EBNA-LP, and EBNA3C transcripts were most prevalent. EBNA2 was focally detected at low level but was still significantly higher in infected than in uninfected gastric cancers. Prior histochemical work has generally not revealed protein-level expression of the EBNAs or lytic viral gene products, so further work is required to learn if these virally encoding RNAs are localized to malignant cells, lymphocytes, or possibly even to exosomes or virions in the extracellular milieu.
Compared to uninfected cancers, the infected cancers had significant upregulation of nine cellular factors (FCER2, MS4A1 (CD20), PLUNC, TNFSF9, TRAF1, CXCL11, IFITM1, PPARG, and FCRL3), implying that EBV is not an innocent bystander with respect to biochemical impact. The virus-associated changes we found were in pathways known to viral oncologists, namely NFKB and NOTCH signaling (FCER2, TRAF1, PPARG) and mucosal immune response (PLUNC, TNFSF9, CXCL11, IFITM1, FCRL3). MS4A1 (CD20) is B cell specific, reminding us that some of the factors upregulated in EBV-infected compared to uninfected gastric cancers could derive from stromal elements rather than from malignant epithelial cells. PLUNC was previously described as a tumor marker for gastric and nasopharyngeal carcinomas, and it encodes a secreted protein involved in innate immune response
[55–57]. TNFSF9, a cytokine of the tumor necrosis factor family, stimulates T cell activation and triggers IFNG production which in turn induces the proinflammatory chemokine CXCL11 and the innate antiviral factor IFITM1. PPARG is as a nuclear receptor controlling glucose metabolism and microtubule networks, and it is a promising target for inhibitory drugs
. The FCRL3 immune response gene is mutated in autoimmune diseases such as rheumatoid arthritis, lupus, and Grave’s disease.
Our findings support the work of Lee et al who found distinct human expression patterns in infected versus uninfected gastric cancers
. Although their study targeted protein and ours targeted RNA, our findings agreed with theirs for 4 of the 5 factors in common between the two studies (BCL2, PTEN, CDH1, PTGS2). There was a potential discrepancy for ERBB2 that was significantly less frequently expressed in infected compared to uninfected gastric cancers when tested at the protein level
, whereas the current study showed no significant difference at the RNA transcript level. Confounding factors include 1) the proportion of tumor cells present in the specimens evaluated, 2) different criteria for categorizing expression status, and 3) RNA versus protein targets.
In general, the array technology that was used in this study worked remarkably well in generating RNA profiles that were believable by virtue of distinguishing known benign versus malignant and gastric versus cervical histopathologies. Furthermore, co-expression of analytes in the same pathway or by the same infectious agent makes sense from a pathobiology and virology perspective. Interestingly, all of the cervical tissues clustered together, and benign and malignant cervical lesions were largely segregated even though the Gastrogenus v1™ test panel had not been specifically designed to achieve these endpoints. Lack of multiple co-expressed EBV mRNAs in cervical tissues reinforced what we knew about their EBV-negativity by the gold standard EBER in situ hybridization assay.
Among the seven genes that were significantly more expressed in gastric cancer (regardless of infection status) compared to lymphoepithelioma-like cervical cancer, four were previously reported as gastric cancer markers (CLDN18, REG4, OLFM4, CDH17)
[55, 59–63]. Two others (EPCAM epithelial cell specific trans-membrane glycoprotein, and PPARG chemokine), as well as REG4, are being explored for targeted cancer therapy
[64–66]. The last of the seven, BBC3 (also called p53 up-regulated modulator of apoptosis, or PUMA) is reportedly upregulated by EBV LMP2A and reigned in by EBV miR-BART5 in cell line models
[67, 68], suggesting that this BCL2 family member is tightly regulated by the virus.
One of the two RNAs that was significantly higher in cervical compared to gastric cancer was IFITM1, which you may recall was also found to be overexpressed in infected compared to uninfected gastric cancers. Further work is needed to explore if cervical cancers (presumably human papillomavirus-infected) and EBV-infected gastric cancers share a common virus-related mechanism for overexpression of this innate immune response factor. The other gene significantly overexpressed in cervical compared with gastric cancer was HIF1A whose expression was associated with that of four downstream angiogenesis mediators in our panel (VEGFA, SLC2A1, SLC2A3 and EPAS1) as evidenced by positive Pearson’s correlation coefficients (data not shown). If confirmed to be operative in vivo, HIF pathway stimulation implies that angiogenesis inhibitors are worth investigating.
Benign versus malignant gastric tissues tend to cluster separately on the heat map, with some exceptions. Field effect
 or exosomal transfer of factors to adjacent regions of the local environment
[70, 71] could explain why some cancers and adjacent reactive tissues had similar profiles. While macrodissection was used to carefully separate benign from malignant lesions, we cannot exclude occult malignancy as a contributor to aberrant clustering.
Among the 19 genes significantly upregulated in gastric cancer compared to adjacent non-malignant gastric mucosa, most were previously reported as gastric cancer specific markers
[72–76], and we now confirm that their upregulation is detectable in archival paraffin-embedded tissue. Lower levels of GAST (gastrin) RNA in cancer tissues could help explain the concomitant loss of the gastrin signaling factor CHGA (chromogranin). The most consistently downregulated factor in gastric cancer versus adjacent benign mucosa was the tumor suppressor gene CDH1 (E-cadherin) suggesting either 1) CDH1 promoter hypermethylation
, 2) rare germline mutation of CDH1 associated with heritable predisposition to gastric cancer
, or 3) downregulation of CDH1 by EBV LMP1 as described in cell line models
LMP1 was previously reported to be absent in infected gastric cancer except in rare cases
[50, 51, 80, 81]. It was therefore surprising that Nanostring nCounter array profiling showed consistent albeit low level expression of LMP1 RNA along with virtually all of the other EBV RNAs that were tested in the infected gastric cancers. Coordinated co-expression of multiple viral genes argues that the expression is true positive. Our microarray results raise the possibility that the viral RNAs we detected are not encoding proteins or that the proteins are 1) only transiently expressed, 2) rapidly degraded, 3) localized to rare cells that are promptly recognized and destroyed by the immune system, or 4) present at such low level that traditional assays are too insensitive to detect them
. The nCounter test system manufacturer claims analytic sensitivity equivalent to that of rtPCR
While most viral genes were expressed almost exclusively in the infected gastric cancer cohort, EBER1 and EBER2 were commonly expressed in each one of the benign and malignant gastric and cervical cohorts, albeit at much lower levels than was seen in each of the EBV-infected gastric cancers. Indeed, our study revealed a novel way to identify EBV-infected gastric cancer by measuring EBER1 and/or EBER2 RNA in archival tissue, and we have proposed thresholds that successfully distinguish infected from uninfected gastric cancer.
Support for active viral infection in infected gastric cancer patients comes from serologic evidence of higher titers against viral capsid antigen compared to EBV-negative gastric cancer patients and benign controls
. Low level lytic infection was previously described in mucosal lymphoid cells
[31, 82, 84] and in infected gastric epithelial cell lines
. BARF1 is known to be expressed in gastric cancer where it is proposed to act as a latent rather than a lytic factor
[50, 51]. Using sensitive rtPCR technology, multiple EBV lytic transcripts were detected by Luo et al in gastric cancer tissues
. Whether active replicative infection occurs in malignant epithelial cells or in lymphoid cells remains uncertain since histochemical stains have failed to reveal a cellular source of lytic factors in gastric tissues
While EBV-infected gastric cancer is biologically distinct from EBV-negative cancer in some respects, the infected counterparts still share many of the classic features previously identified as being characteristic of gastric cancer, such as specific collagens (COL1A1, COL1A2, COL3A1), SULF1, THY1, SPP1, INHBA, and SPARC. These pan-gastric cancer markers might be exploited for early diagnosis or for monitoring tumor burden during therapy, especially when multiple such markers are tested in concert to maximize specificity while still capturing the heterogeneity of the disease. Biomarkers for the EBV-infected subset, such as EBV DNA and the highly expressed viral EBER1
EBNA1, and BRLF1 RNAs, as well as associated cellular factors confirmed in this study, represent promising targets for early detection. To the extent that any of these factors circulate in blood, they might serve as non-invasive indicators of disease analogous to what has already been achieved for two other EBV-infected neoplasms-- post transplant lymphoproliferative disorder and nasopharyngeal carcinoma. In both of these disorders, Q-PCR of circulating EBV DNA facilitates early diagnosis and in monitoring efficacy of therapy
[86–88]. High levels of EBER1 and EBER2 RNA were measurable in plasma of 89% of nasopharyngeal carcinoma patients
Antiviral therapy is becoming more accepted given its biologic underpinnings-- the viral genome is present in every malignant cell of a given infected cancer-- thus making the virus one of the most appealing therapeutic targets in our armamentarium. Off-the-shelf cytotoxic T cells are now available to treat selected EBV-related malignancies
[90, 91]. Early clinical trial data demonstrate the merits of lytic induction therapy
[33, 92, 93]. Assessment of lytic induction by panels of tests such as the microarray system described herein could be useful for measuring the biochemical impact of an intervention and its efficacy.
Applicability of the Nanostring nCounter system to archival paraffin embedded tissue was previously reported by others
[43, 44], but ours is the first study to examine viral and human RNAs in concert. The test system’s ability to rapidly profile multiple RNAs generates rich data relevant to viral oncology and patient care. A major advantage is suitability for routine fixed tissue specimens including small biopsies that were previously collected, processed and stored using customary clinical methods. While microscopy is essential to assuring that representative tissue is input into the assay, the noteworthy flexibility of the test system with regard to malignant cell proportion promotes it use in clinical settings. Panels of analytes could be tailored to support different intended uses such as suitability of a subject for a specific clinical trial, or monitoring efficacy of a given regimen in serial specimens.