In this paper we have reported on the expression profiling of EBV miRNAs in a wide variety of infected normal and neoplastic tissues that express all of the known EBV associated latency transcription programs. We have shown that there are distinct patterns of miRNA expression associated with Latency III and the restricted forms of latency (Latency II/I/0) and that these patterns are deregulated in EBV associated tumors. GCB (Latency II) and MemB (Latency I/0) express the same unique, restricted pattern of miRNAs with the exception that mirBART 17-5p is preferentially expressed in MemB cells. This unique pattern includes the absence of 12 BART miRNAs that include approximately half of Cluster 2. At least 8 of these absent miRNAs are expressed in B cells expressing Latency III (LCL) and in all of the tumor biopsies we have tested although none of the tumors uses Latency III. Thus this is a Latency III restricted pattern of miRNA expression, which we refer to as Latency III associated BARTs, that is deregulated in tumors. Interestingly another group of miRNAs that are associated with Latency III, the BHRF1s 
, are not deregulated in the tumors suggesting that viral miRNA deregulation in the tumors is specifically targeted at the BARTs. This is an important conclusion since it represents the first demonstration of Latency III specific gene expression in tumors that are otherwise expressing restricted (Latency I/II) forms of latency
and raises the possibility that BART miRNAs may contribute to oncogenesis. The observation of latency program specific miRNA expression could only have been made by studying in vivo derived infected material since the Latency III associated BARTs are expressed in all tumor biopsies and cell lines we have tested.
In the generally accepted model of EBV persistence, newly infected naïve B lymphoblasts (LCL) expressing Latency III, switch to Latency II when transiting the germinal center (GCB) to become resting memory B (MemB) cells, the site of long term persistence where viral latent protein expression is extinguished 
. Our results here suggest that the transit from EBV driven growth into more restricted forms of latency in GCB and MemB cells is associated with turning off expression of the Latency III associated BARTs and up regulation of the remaining BART miRNAs by 5–10 fold. This is not simply related to the cessation of proliferation because latently infected GCB are, like LCL cells, proliferating 
. Since proliferation in the GC is not driven by Latency III, the viral growth program, we may conclude that the Latency III BART profile (presence of the Latency III associated BARTs and reduced expression of the remaining BARTs) is specifically associated with EBV driven growth. One caveat to this conclusion is that it would have been desirable to confirm the Latency III pattern of miRNA expression on in vivo infected cells rather than spontaneous LCL. Unfortunately newly infected tonsil naïve B cells in vivo are present at a level 5–10 fold lower than infected GCB 
. This puts them below the threshold of sensitivity and reliability for our profiling and therefore was technically not feasible.
Several studies have reported on potential roles for EBV miRNAs. There is no striking correlation between these reports and the BART miRNAs we have as the Latency III associated BARTs. Marquitz et al 
have suggested that Cluster 1 and 2 BART miRNAs interact in apoptosis resistance by targeting BIM. However, their observations are not consistent with those of Seto et al, who have reported that BART miRNAs have no impact on LCL growth or survival in vitro
, or that the entire BART region can be deleted without impacting the transforming capacity of the virus
or that the prototypical laboratory strain B95-8 has most of the BART region, including most of Cluster 2, deleted yet is unimpaired in its transforming ability. The explanation for this discrepancy may lie in the fact that Marquitz et al performed their studies in an epithelial cell line not in B cells. Taken together these results suggest that the Latency III associated miRNAs we have identified, may play a crucial survival role in vivo for newly infected naïve B lymphoblasts activated by the EBV Latency III program but that this role is dispensable for in vitro growth much as has been shown for LMP2 
. We assume that the specific up regulation of this group of miRNAs in tumors implies they could play a similar survival role in tumor development.
Our results suggest that expression of the Latency III associated BARTS is coordinately regulated. It seems unlikely that this is occurring at the level of transcription/splicing. It is known that the BART miRNAs are derived from the first four introns of the BART transcript prior to the splicing event 
and the miRNAs absent from GCB and MemB cells are not contiguous but randomly distributed among these introns. For examples, Bart 15 is located in the region between exon 1a and 1b whereas Bart 10 and Bart 20 are in the junction of exon 2 and 3. Therefore, it is unlikely that the differential miRNA expression we have described is related to the selection of splicing patterns. Other possible mechanisms that are known to regulate miRNA expression are differential DNA methylation 
and RNA editing both of which have been shown to function on BART miRNAs 
. However, these mechanisms defer rather than answer the question as to why or how this particular subset of miRNAs is targeted for coordinate expression. A mechanisms that we favor is based on the observation that the stability of miRNAs is dependent on the presence of their target mRNA 
. In this case the absence of miRNAs in GCB and MemB that are present in LCL and the tumors would arise because the mRNAs targeted by those miRNAs were only present in LCL and the cancers.
We were surprised to find that the four tumor types clustered together in the heat map. This was irrespective of the tumor type, tissue of origin or the EBV transcription program that they employed. Perhaps more unexpected was our finding that despite the very similar patterns of miRNA expression the different tumor types were nevertheless clearly distinguished in two separate assays (heat map/clustering and PCA) applied to two completely separate data sets (p in both cases
0.001). The basis for this resolution is less clear since it is not associated with any particular subset of miRNAs. Rather we discovered that a majority of all subsets of five or more miRNAs and some subsets as small as three were capable of distinguishing all four cancer types. Even more curious was the finding that every miRNA is capable of contributing to one of these subsets. Taken together, these results mean that the signal distinguishing these cancers is highly redundantly encoded across the miRNA expression profiles. Such a distribution of information is uncommon indeed may never have been reported before for a biological system. However, this type of behavior is well known in physics and computer science where there is a close analogy to secret sharing algorithms 
. For example, it is possible to share a secret message among any number (in our case ~40) people in such a way that if any 5 of them divulge their information to each other, the message can be read. It is interesting to speculate that the signal in our assays, i.e., the causes of or responses to each cancer type is in some similar manner parceled out among the miRNAs. How this might work at the molecular level is unclear but we assume it must reflect extensive redundancy in the miRNAs both in the number that target the same gene and in targeting genes that lie in the same or parallel signaling pathways related to tumor development. Evidence for such layers of redundancy in miRNA function is well known 
and has recently been reported for both the Cluster 1 and 2 BART miRNAs in reducing apoptosis susceptibility in an epithelial cell line 
It has been suggested previously, based on cell lines, that the copy number of BART miRNAs is higher in epithelial cells than B cells. This is consistent with our estimates of the relative abundance of the miRNAs in our tumor biopsy samples versus the LCL. However, the correlation of high expression with epithelial tissue is confounded by our measurement of relative abundance in the MemB and GCB populations where the expressed BART miRNAs are present at comparable levels to the epithelial tumors. The exact meaning of these variable levels of expression is therefore now unclear.
The only EBV latency transcription program that we found to be associated with a specific pattern of miRNA expression was Latency III characterized by up regulation of the Latency III associated BART and BHRF1 miRNAs. The later serves as validation for our approach since it has been reported 
and confirmed 
previously. We found BHRF1 expression in some of the tumor biopsies notably BL and HD although they were present at a very low level which we estimate to be generally less than one copy per cell. The absence of BHRF1 miRNAs from EBV associated tumors is consistent with previous findings that BHRF1 miRNAs were not found in biopsies from GaCa 
and DLBCL tissues 
. By contrast we found abundant BHRF1 miRNAs in the tumor derived cell lines. This, together with our finding that the BART miRNA expression profile in these lines tended to resemble LCL more closely than the originating tumor biopsies, casts doubt on the value of using such lines to study EBV miRNAs.
In conclusion, we have presented the first comprehensive profiling of EBV miRNAs from in vivo derived normal and neoplastic tissue. These results demonstrate specific patterns of expression in Latency III versus more restricted forms of latency and deregulation of miRNA expression in tumors.