Viruses have been shown to induce tumor formation through numerous and diverse mechanisms (5). From the pioneering studies of Peyton Rous in 1920s, retroviruses were found to encode oncogenes, like v-src, v-myc, v-ras, which could potently transform normal cells to grow as tumors in animal models (6). Several of these viral oncogenes were subsequently recognized as host genes captured by viral recombination and their oncogenicity due to combinations of activating mutations and deregulated expression in infected cancer cells. DNA tumor viruses were found to encode novel oncoproteins, such as SV40 T-antigen, adenovirus E1A and E1B, and papillomavirus E6 and E7 that interact with and disable cellular tumor suppressor proteins, such as p53 and Rb (7). While these viral oncogenes may be necessary for transformation in laboratory models, they are not sufficient for tumor formation in the overwhelming majority of natural infections. In addition, some viruses, such as adenovirus, encode potent inhibitors of cellular tumor suppressors p53 and Rb, but rarely associate with human cancers (8). This is consistent with the observation that most oncogenic viruses typically cause benign infections, that only rarely lead to cancer.

Pioneering studies identified a few cellular targets universally exploited by tumor viruses, such as p53 DNA damage surveillance and Rb cell cycle control. However, it is now known that subversion of these targets are not sufficient for viral tumorigenesis, and that additional and alternative host proteins and pathways are targeted by viral oncogenes (Figure 1). For example, the small viral oncoprotein E7 is well known to bind and degrade Rb (9), but can also interact with the Rb-associated DREAM complex (10), phosphatase PTPN1(11), histone modifying enzyme HDACs (12), stem cell promoting factors APH1B and OCT4 (13, 14), and Cullin2 to stabilize APOBEC3a (15). In addition, E7 can cooperate with another viral oncoprotein E6 to activate hTERT to overcome replicative senescence (16, 17). Viral subtypes, as well as host cell type can determine whether these different interactions are oncogenic, further demonstrating the diversity of targets for one small viral oncoprotein and its potential effects on different cancer pathways. Similarly, the MCPyV small T antigen can interact with MYCL and EP400 to alter chromatin and transcription regulatory networks implicated in cell lineage control (18), 4EBP1 affecting translational control (19), protein phosphatase 2A to affect ubiquitin ligases (20), the F-box proteins FBW7 to activate NF-kB signaling (21, 22), as well as binding iron-sulfur clusters (23). This promiscuous multitasking is likely to be a general feature of viral oncoproteins that target multiple cellular proteins and pathways.

Larger DNA tumor viruses, such as EBV and KSHV, encode numerous viral genes implicated in oncogenesis. Many of these oncogenes target pathways important for the tissue-specific functions of the host cell. EBV encodes two membrane oncoproteins, LMP1 and LMP2, that cooperate to immortalize primary B-lymphocytes by mimicking B-cell receptor and CD40 co-receptor (24, 25). LMP1 interacts with multiple TRAFs and TRADDs to activate NF-kB pathways (26), but can also interact with other proteins involved in membrane vesicle formation (27). LMP2 can interact with several different src-family kinases (28). EBV also encodes 6 nuclear antigens, EBNA-LP, 1, 2, 3a, 3b, and 3C, that are all implicated in oncogenic mechanisms (29). EBNA2 is absolutely required for B-cell immortalization in vitro (30), and natural mutations in EBNA2 correlate with B-cell transformation activity (31). However, some EBV associated tumors fail to consistently express EBNAs and LMPs. Like EBV, KSHV also encodes candidate oncogenes, including the nuclear antigen LANA that can bind host chromatin and alter p53 and Rb function, vGPCR that can induce endothelial tumors in transgenic mouse models, vCyclin that can drive cell cycle progression, vFLIP and K12 that can activate NF-kB and STAT3 signaling (32). In addition to these viral encoded proteins, both EBV and KSHV have numerous non-coding RNAs implicated in oncogenesis. EBV small non-coding RNA EBERs can interact with transcription factor PAX5 (33), ribosomal protein L22 (34), TLR receptors (35), and provide paracrine signals through exosome transmission (36). Numerous viral miRNAs target oncogenic pathways implicated in EBV and KSHV carcinogenesis (37, 38). EBV miRNA are highly expressed in tumors, especially EBV-associated gastric carcinomas (EBVaGC) along with other non-coding RNAs that arise from the same genomic locations (BARTs) and have additional oncogenic potential (39, 40). Remarkably, these viral genes are expressed at variable levels and heterogeneously in most viral-associated cancers, further confounding the problem of complexity and diversity of viral oncogenic mechanisms.

All known human tumor viruses persist, in one form or another, at the site of tumor formation. Tumor viruses can persist as chronic infections (HCV), nuclear episomes (HBV, HPV, MCPyV, EBV, KSHV) or integrated genomes in viral-associated tumor cells (HTLV-1). HCV is an unusual oncovirus in that it does not infect the cancer cell, but its long-term persistence causes inflammation conducive to cancer cell emergence. In contrast, episomal DNA tumor viruses, like EBV, KSHV and HPV have viral-specific programs dedicated to viral genome persistence in a dividing cell that can serve as a clonal outgrowth in cancer (119–121). These viruses encode proteins, such as EBNA1, LANA, and E2, dedicated to binding viral DNA and maintaining the viral genome over generations in proliferating cells. HTLV-1 persists through integration into the host genome as part of its normal life cycle (122). Integration is inherently mutagenic and there is evidence that some, albeit rare integrations are oncogenic.

Chang and Moore proposed that viral cancers arise due to aberrations in the normal productive life cycle, including genetic mutations and integrations that disrupt normal viral gene expression (123). Consistent with this, aberrant integrations are frequently observed for oncoviruses that typically persist as extrachromosomal episomes, such as HBV, HPV, and MCPyV. HBV has been found to integrate in oncogene hot-spots, such as the TERT (telomerase) or KMT2B (MLL4) loci (124), and HPV has been found to integrate at ERBB2 and PTPN13 loci (125). In each case, integration alters normal gene regulation to promote oncogenesis. Integrations can also lead to loss of viral DNA and deregulation of viral oncogenes. HPV integration with loss of viral E2 repressor protein leads to the upregulation of viral oncogenes E6 and E7 in cervical carcinomas (126, 127). In Merkel cell carcinomas, MCPyV frequently integrates as incomplete genomes with deletions in large T and overexpression of small T (18, 128). There is some evidence that viruses can transform host cells without viral genome persistence through heritable changes in the host epigenome or regulome (129). Such “hit and run” mechanisms have been reported but are difficult to demonstrate in naturally occurring human cancers. Thus, most viral cancers are associated with long-term persistence of viral genetic material transmitted for multiple cellular generations.

Tumor cell heterogeneity is a central hallmark of cancer (130–134). Viruses can be highly heterogeneous, as well as induce a more heterogeneous phenotype in the host cells. Viral gene programs are inherently unstable with potential shifts from productive to non-productive, or latent to lytic infection cycles with variable gene expression patterns. Viral genomes may be more relaxed than host chromatin and free to move from one chromosome compartment to another. Viral genes may be activated or repressed with greater flexibility, including sporadic bursts of lytic amplification and gene expression. Oncoviruses can also increase variability in host gene expression. Viral infected tumor cells maintain a poorly differentiated state, and may toggle between the lympho-epithelial features of EBV NPC and GC, and the mesenchymal-endothelial features of KSHV infected KS spindle cells (135). This loss of fixed cell identity has been referred to as cellular plasticity. Increasing cellular plasticity provides cancer cells with the advantage of increase variation, or the capacity to adapt more rapidly to changing environmental conditions (such as hypoxia), relocate to new niches (metastasize), evade immune surveillance, and develop drug resistance (136).

Viruses can be highly heterogenous and mutate during the course of infection to increase carcinogenic risk. High and low risk subtypes of HPV may be considered a form of species heterogeneity, while integrations and deletions can be a source of mutational variation (123, 137). Tumor heterogeneity based on viral and host gene expression (including single cell RNAseq), epigenetic modifications, and immune infiltration have been observed for cancers associated with HPV (138), MCPyV (139), HBV (140), HCV (141), and HTLV-1 (142), indicating that such heterogeneity is a general rule for viral and non-viral cancers. Variations in viral gene expression and genome copy number may also account for cancer-risk. For EBV, viral latency types can contribute to genetic heterogeneity and plasticity. EBV can adopt different latency types in different host cells and tumor types. Epigenetic factors are known to regulate the different latency types, including differences in DNA methylation patterning. Among the viral genes with variable expression is the potent oncogene LMP1, involved in constitutive TNF-pathway signaling. LMP1 gene can be expressed at ranges that vary 100 fold among single cells in a population (143). Variations in LMP1 expression in NPC correlated with cellular genes linked to E2F and cellular DNA replication, as well as to changes in NF-kB and JAK/STAT signaling (144). Another variable in viral gene expression is the genome copy number (145). Viral episomes can range from 1 to several 100 copies per cell. Viral genome amplification is closely linked with over-expression of some viral genes, especially the lytic cycle genes of the gammaherpesviruses. Although a complete lytic cycle is rarely detected in viral tumors, abortive lytic cycle gene expression is likely to contribute to viral carcinogenesis. These lytic cycle genes provide numerous potential contributions to viral carcinogenesis, including anti-apoptotic, immunomodulatory, and paracrine activities required for tumorigenesis (146).

Temporal history of viral infection may also contribute to tumor heterogeneity (Figure 2). Viral cancers are often clonal expansions of a virally infected progenitor cell (147, 148). However, the progenitor cell is likely to have acquired somatic mutations or other aberrations enabling viral transformation. Pre-existing somatic cell mutations appear to be required for the formation of EBV-associated NPC (149). EBV infected nasopharyngeal cells undergo cell cycle arrest, unless infected cells have pre-existing mutations in cell cycle control. Gene loss from 9p21 to 3p21.3 (inactivating RASSF1A and CDKN2A1, respectively), and activation of telomerase, have been found to precede acquisition of EBV infection in the evolution of NPC (150). EBV infection further drives clonal expansion of infected cells, as has been demonstrated in NPC by examining the uniformity of the viral terminal repeat DNA (148). Subsequent genomic hypermethylation follows the infection of EBV, and then activation of NF-kB pathways, loss of MHC I, mutations in PI3K/MAPK, and chromatin remodeling, and subsequently TP53 and RAS, along with other mutations (151) are frequently observed in the course of NPC formation (Figure 3).

Host genetic variation, whether inherited or acquired through somatic mutation, also contribute to the risk of viral cancers. Variations in the HLA locus correlate with risk of viral cancers, suggesting that presentation of viral antigens plays a key role in immune resistance to viral-driven cancers. Genetic analysis of NPC susceptibility revealed risk loci at hTERT, CDKN2A/B, MECOM, and TNFRSF19, all of which have known roles in oncogenic pathways. Other susceptibility pathways have been linked to Notch signaling, magnesium transport (NIPAL1), EBV entry into epithelial cells (ITGB6), modulation of apoptosis (NEDD4L, BCL2L12), cAMP signaling, or DNA repair (MLH1, PRKDC) (153). Inherited mutations in magnesium channel MAGT, as found in XMEN syndrome is associated with defects in NK and T-cell control of EBV infected B-cell (154, 155), and may also contribute to risk of NPC and KSHV associated KS.

Other aberrations also contribute to variations in viral-host interaction. Rare tumors of atypical tissue types, such as EBV leiomyosarcomas, NK-T cell lymphoma, peripheral T-cell lymphoma (156) and pulmonary lympho-epithelium-like carcinoma (LELC) (157) are likely due to aberrant entry of EBV into unnatural host cells. Environmental factors, such as coinfection with HIV or malaria, can alter the immune control of viral infected cells. Thus, host cell type, immune functionality and other environmental factors can impact the course of infection and cancer progression in a temporal-dependent manner.

Stage specific effects on viral oncogenesis are observed for most oncoviruses. HPV has been shown to have different patterns of infection, integration, and gene expression at different stages of viral-associated cancers (158). Similarly, MCPyV T-antigen was found to have stage specific tumor promoting activity in a mouse model treated with defined carcinogens (159). HTLV-1 antigen expression changes in response to T-cell activity, providing evidence for viral adaptation and co-evolution with tumor cell progression (160). HBV X-protein interaction with miRNA production impacts multiple stages of HCC through very different pathways, ranging from cell cycle control at early stages to immune suppression at later stages (161). These multifunctional viral oncoproteins can affect stage specific events in cancer cell evolution, and therefore may adapt their oncogenic activities activities over time.

Oncogenic viruses increase the number of possible ways a cell can propagate, survive and achieve oncogenic transformation. In terms of Darwinian evolutionary dynamics and population genetics, virus infection increases the phenotypic diversity and fitness of the population. A more heterogeneous population has a fitness advantage by adapting more rapidly to changing and stressed environments. To borrow from systems biology, a stable phenotype requires a stable gene regulatory network (GRN). GRNs are considered a thermodynamic “attractor” state (162) (Figure 4). GRNs are related to the developmental states described by Waddington (164) using an energy landscape and the canalization patterns that separate two or more distinct cell fates or GRNs. It has been proposed that cancer cells converge on a common GRN attractor state akin to the embryonic and unicellular cell states (165). Oncogenic viruses perturb major hubs in GRNs enabling greater plasticity between phenotypic states (136). Experimental validation of this concept has been provided by measuring the intrinsic plasticity of EBV positive Burkitt lymphoma cell lines (162). Viruses can also increase signal noise in a GRN [reviewed in (166)]. Viral genomes may have inherently higher “noise” than their cellular counterparts due, in part, to their relaxed epigenetic regulation, subcellular localizations, and copy number variations. Viral genomes and gene products destabilize GRNs and facilitate the transition from one attractor state to another (167). Thus, we propose that a major feature of oncogenic viruses is their ability to accelerate the rates of cancer cell evolution by increasing the genetic variability and phenotypic plasticity, and inherent cellular adaptability to changing and stressful microenvironments (168, 169).

In thermodynamics and statistical mechanics, terms like entropy were developed to explain the behavior of complex systems with excessively large numbers (ensembles) of microstates. In this respect, the term entropy can be used to describe the number of microstates of a complex biological system, such as the gene regulatory interactions in a viral infected tumor cell. While we can not provide a rigorous definition of biological entropy, we do suggest that viruses increase the number of possible microstates available to the host cell, and therefore may be considered a form of “genetic entropy.” In the most simplistic terms, genetic entropy may be the ability to reconfigure the genome and its programmed processes. Viruses reconfigure genomes, gene expression programs, and biochemical pathways. We further suggest that this be considered in terms of Shannon information theory, where viral genomes may be considered a source of signal noise enabling the freedom to find a lower energy state, or alternative GRN, such as the oncogenic state. Viruses can increase signal noise and alternative outputs for cellular developmental programs, and this enables transcriptional plasticity and phenotype heterogeneity associated with tumorigenesis. Efforts to quantify cellular information and thermodynamic entropy may be useful for understanding the emergence of the cancer phenotype. In this regard, virus infection may be considered an entropic driving force for cancer (170).

The diversity of oncogenic mechanisms and the plasticity of cancer cells raise enormous challenges in developing precision therapies. For most cancers, early detection provides the best opportunity for effective treatment. Viral cancers have the advantage of having viral-specific targets and biomarkers. For example, early-stage NPC can be predicted from cell-free EBV DNA in plasma (171) and EBV-specific IgA (172) and can be effectively treated with radiation. However, most cancers are discovered at later stages and fail treatment or develop resistance and recurrence. This is largely attributed to tumor heterogeneity and plasticity and the emergence of resistant clonal populations. Oncogenic viruses provide a rich resource for tumor cell heterogeneity and evolutionary diversity. Elimination of persistent oncogenic viruses at early stages is ideal, but not always possible. Reduction of genetic plasticity and modulation of selection pressure may be attractive alternative approaches for treatment of viral cancers. It may also be possible to use evolutionary principles to improve dosing and timing of therapy. Additionally, it may be possible to exploit viral genetic plasticity to eliminate viral cancers. Since excessive genetic variability can be incompatible with life (173), it may be possible to amplify viral-induced plasticity through drug intervention. Treatments that increase gene regulatory noise, such as epigenetic modifiers, could provoke chaotic and lethal gene expression patterns in cancer cells that would be resisted by normal cells.