The scope of viral causation of human cancers: interpreting virus density from an evolutionary perspective

Abstract

Most known oncogenic viruses of humans use DNA as their genomic material. Research over the past quarter century has revealed that their oncogenicity results largely from direct interference with barriers to oncogenesis. In contrast to viruses that have been accepted causes of particular cancers, candidate viral causes tend to have fewer viral than cellular genomes in the tumours. These low viral loads have caused researchers to conclude that the associated viruses are not primary causes of the associated cancers. Consideration of differential survival, reproduction and infiltration of cells in a tumour suggest, however, that viral loads could be low even when viruses are primary causes of cancer. Resolution of this issue has important implications for human health because medical research tends to be effective at preventing and controlling infectious diseases. Mathematical models may clarify the problem and help guide future research by assessing whether low viral loads are likely outcomes of the differential survival, reproduction, and infiltration of cells in a tumour and, more generally, the extent to which viruses contribute to cancer.

This article is part of the theme issue ‘Silent cancer agents: multi-disciplinary modelling of human DNA oncoviruses’.

1. Introduction: evolution and selection in viral oncogenesis

Three processes of evolutionary selection are central to oncogenesis, the development of normal cells toward and beyond the threshold of cancer. The first, oncogenic selection, is analogous to natural selection because it generates an evolutionary change in a living system, but differs because the change results from elevated survival and reproduction of aberrant cells within an organism rather than differences in survival and reproduction of organisms [1–3]. The second process is the action of natural selection on multicellular organisms to reduce their risks of cancer [4,5]. The third process is the action of natural selection on infectious agents, particularly viruses, that generate oncogenic effects [6,7]. A coherent understanding of viral causation of cancer requires the integration of these three processes.

(a). Oncogenic selection

Normal cells evolve into cancer cells in part through selection acting on genetic variation generated by mutations. The difference between oncogenic selection and natural selection is manifested in two ways. The first involves differences in long-term opportunities for evolutionary adaptation. Natural selection acting on an organism can result in cumulative changes over unlimited numbers of generations, whereas oncogenic selection of somatic cells within an organism is truncated by the death of the organism. Oncogenic selection, therefore, cannot generate the degree of biological innovation that arises from natural selection.

The second difference involves the tendency for oncogenic selection to dysregulate rather than perfect adaptations that have been generated by natural selection. Oncogenic selection involves the differential survival and reproduction of cells within the organism rather than heritable changes in the genetic composition of organisms within a population. In response to natural selection, cells in multicellular organisms generally restrict their own survival and reproduction. Oncogenic selection favours cells that break rather than refine these restrictions.

(i). Adaptations that protect against cancer

The barrier theory of oncogenesis explains causes of cancer in the context of barriers and restraints. Barriers are defined as adaptations that block oncogenesis when they are in place. Restraints are defined as processes that retard but do not block oncogenesis [5]. The identified barriers to oncogenesis are cell-cycle arrest, apoptosis, telomerase regulation for non-stem cells, cell adhesion for metastatic cancer and asymmetric division for stem cell cancers [5]. Restraints on oncogenesis are much more numerous and include restrictions of resources, vulnerability to immunological defences and regulation of the division rate of a dividing cell. Abrogation of barriers and restraints are considered essential and exacerbating causes of oncogenesis, respectively.

This barrier theory of cancer is similar to but distinct from the hallmarks of cancer [8,9], which has been one of the central conceptual frameworks for organizing knowledge about cancer, including the relationships between oncogenic viruses and cancer [10]. A fundamental difference between the two conceptual frameworks is that the barrier theory attempts to distinguish essential from exacerbating causes of cancer, whereas the hallmarks of cancer identify the major features of cancers. Abrogations of barriers are important features of cancer but not all of the hallmarks involve the abrogation of barriers. The barriers theory thus identifies a subset of the hallmarks as processes that interfere with barriers to cancer and are therefore essential causes. For the remaining hallmarks, current information either indicates that a barrier is not broken (e.g. induction of angiogenesis, deregulating cellular energetics, promotion of inflammation) or is insufficient to determine whether the hallmark is an essential cause (e.g. immune avoidance, genomic instability). Another distinction involves asymmetric cell division, which is considered a barrier for cancers derived from genuine stem cells (as opposed to ‘cancer stem cells’) even though abrogation of asymmetric division is not considered a hallmark of cancer.

These distinctions can be important for distinguishing bona fide oncogenic viruses from viruses that may contribute more rarely and haphazardly to oncogenesis, for example, by causing inflammation. A virus may regularly abrogate a barrier but rarely if ever contribute to oncogenesis because other barriers remain functional. Inhibition of apoptosis, for example, will generally be favoured by natural selection acting on viruses because apoptosis terminates survival of viruses within the apoptotic cell. The human papillomavirus (HPV) types 6 and 11, for example, are not considered to be oncogenic viruses but suppress apoptosis [11]. Inhibition of cell-cycle arrest also has been reported for HPV11 [12]. Such ‘non-oncogenic’ HPVs can contribute to oncogenesis under particular environmental conditions, such as when vulnerability alleles and sunlight exposure lead to epidermodysplasia verruciformis and associated skin cancer. The barrier theory offers a resolution to such ambiguities by providing a basis for ascribing the label ‘oncogenic virus’, namely that the virus needs to compromise all of the barriers to cancer that are present in the particular type of cell that is infected. In the case of Epidermodysplasia verruciformis, HPV11 can be considered oncogenic in the context of mutagenic exposure to ultraviolet light, a genetic vulnerability and immune suppression, even though experts generally do not refer to this and other HPV types associated with warts as oncogenic viruses [11,13].

(ii). Natural selection on viruses favouring oncogenic tendencies

Parasites, mostly viruses, play a causal role in approximately 20% of human cancer [14]. The extent to which viruses contribute to the remaining 80% of human cancer is not known, because infectious causes can be ruled out for very few of these cancers. The recognized oncogenic viruses are mostly DNA viruses (table 1, column 2). Each is also suspected of contributing to other cancers of uncertain cause (table 1, column 4).

Table 1.

Accepted and candidate viral causes of human cancer.

		cancers for which the virus is:
virusa	viral genome	an accepted cause	a candidate cause
EBV	DNA	Burkitt's lymphoma, Hodgkin's lymphoma, gastric carcinoma, post-transplant lymphoma, nasopharyngeal carcinoma	breast, acute lymphoblastic leukaemia, ovarian, lung
HPV	DNA	cervical, oropharyngeal, penile, anal, vulval, vaginal cancers	breast, bladder, oesophagus, prostate, lung, skin
HTLV1	virion RNA with integrated DNA	adult T-cell leukaemia and lymphoma	none
HHV8	DNA	Kaposi's sarcoma	lung
HBV	DNA with RNA intermediate	hepatocellular carcinoma, cholangiocarcinoma	pancreas
HCV	RNA	hepatocellular carcinoma, cholangiocarcinoma	pancreas, non-Hodgkins lymphoma, oropharyngeal
MCPyV	DNA	Merkel cell carcinoma	lung

^aEBV, Epstein Barr virus, human herpes virus 4; HPV, human papilloma virus; HTLV1, human T lymphotropic virus 1, human T-cell leukaemia/lymphoma virus 1; HHV8, human herpes virus 8, Kaposi's sarcoma-associated herpes virus; HBV, hepatitis B virus; HCV, hepatitis C virus; MCPyV, Merkel cell polyomavirus, human polyomavirus 5. Adapted from Ewald & Swain Ewald [15]. For references, see Ewald & Swain Ewald [15].

Viruses may contribute to cancer directly by interfering with barriers and restraints, or indirectly by increasing the frequency of genetic mutations, particularly nucleotide alterations, chromosomal rearrangements and insertional mutagenesis. During the early 1960s, before viruses were accepted causes of human cancer, mutational, epigenetic and metabolic effects of viruses were considered as possible mechanisms, acting separately or jointly [16,17]. Considering the role of viruses, Salvador Luria [18, p. 678] wrote, ‘the relationship of the virus to the evolution of the tumor cell may include a variety of alternatives. At one extreme, the virus may master-mind the whole process; at the other extreme, it may trigger only a single step.’ By the late 1970s, the emphasis on mutational contributions to cancer (e.g. [1]) led to the sense that viruses were contributing to human cancer much like other carcinogens, through generation of mutations and other insults that haphazardly result in destructive proliferation by disrupting cellular regulation (e.g. [19]). Research since then, however, has confirmed viral roles during the early pre-cancer stages of oncogenesis that are closer to the ‘master-mind’ end of Luria's spectrum of possibilities (figure 1). In these virally induced cancers, mutations are generally needed to complete the transformation from proliferative precancerous cell past the threshold of cancer. Infections are still presumed to contribute to oncogenic point mutations through stimulation of damaging compounds during inflammatory processes or by increasing cellular proliferation [20–23], but the major oncogenic effects of human tumour viruses are now known to involve direct interference with barriers and restraints.

Figure 1.

Contrast between viral contributions to oncogenesis. Arrows with dotted lines correspond to the analogy between viruses and environmental carcinogens. Solid lines correspond to Luria's master-mind metaphor for tumour viruses insofar as it relates to the precancerous phase of oncogenesis. Evidence generated since Luria's paper has strongly supported the master-mind hypothesis, though not to the exclusion of indirect viral effects. (Online version in colour.)

Insertional mutagenesis holds an ambiguous place in this conceptual framework. It could directly interfere with oncogenesis if the viral insertion site negates a barrier or restraint. If insertion is random then its contribution to cancer would be more haphazard and analogous to random point mutations. Overall insertional mutagenesis seems to play a relatively minor role. The hepatitis B virus (HBV) infections, for example, can foster cancer through insertional mutagenesis at sites that are relevant to oncogenesis, but these insertions comprise a small percentage of all viral insertions [24,25]. HBV is associated with hepatocellular cancer even when it is not integrated and when it is integrated at sites unrelated to oncogenesis. HBV proteins, however, directly compromise the major barriers to oncogenesis (reviewed by Ewald & Swain Ewald [7]).

(iii). Viral persistence and inconspicuousness of viral oncogenesis

Viral interference with barriers to cancer apparently has been favoured by natural selection because barriers to cancer are also barriers to viral persistence [15]. Disrupting cell-cycle arrest allows the viral genome to replicate. Fostering synthesis of telomerase allows the virus in the infected cell to divide indefinitely. Inhibiting apoptosis reduces the chance that the virus will be destroyed in response to the detection of its presence. Reducing cell adhesion allows the virally infected cell to disperse to new locations in the body. More generally, inducing cellular proliferation allows the viral genome to replicate along with the host cell proliferation without exposing virions to extracellular immunological defences such as antibody attachment.

Even when viruses multiply by stimulating their host cells to replicate, immunological responses impose restraints on oncogenesis. A critical restraint involves the killing of infected cells by cytotoxic T cells. Although oncogenic viruses can overcome this restraint to proliferate and eventually cause cancer, the destruction of infected cells may have an important effect on recognition of the breadth of infectious causation of cancer, because destruction of infected cells may act in concert with several other cellular interactions to lower the occurrence of infected cells in a tumour and thus make viral causation of cancer less apparent.

In this paper, we address this complex set of interactions to help facilitate recognition of the full scope of infectious causation of cancer. We first discuss the direct effects of oncogenic viruses on oncogenesis, using the human herpes virus 4 (Epstein Barr virus, EBV) as an illustrative example. We then focus on the possibility that a low frequency of virus genomes relative to host cells in a tumour (referred to as viral load) may contribute to inconspicuousness of viral causation of cancer and hence to a difficulty in recognizing cases of viral oncogenesis. We then apply this perspective to Hodgkin's lymphoma, to illustrate that viral causation of a cancer has been accepted even though the virus is present at low frequency in the tumour, then to glioblastoma to illustrate how a cancer with low-to-moderate reported loads of human cytomegalovirus is presently in an ambiguous zone between acceptance and rejection of viral causation. We then discuss breast cancer to illustrate how a cancer that has been correlated with five different viruses, often with multiple confirmations and refutations by different laboratories, is not close to acceptance, and how this situation might result from very low viral loads in tumours. Finally, we suggest how these ambiguities may reflect the current state of the expanding recognition of virally caused cancers to include ever more cryptic examples and the potential value of mathematical modelling in this process.

2. Viral master-minds of oncogenicity

EBV illustrates the precision of adaptations that connect enhanced viral persistence with oncogenesis (for detailed overviews for EBV and other oncogenic viruses, see [7,10,26]). EBV breaks cell-cycle arrest by interfering with the retinoblastoma protein through several specific mechanisms. The retinoblastoma protein imposes cell-cycle arrest by binding to the E2F family proteins, thereby prohibiting them from prompting DNA synthesis. EBV's nuclear antigen 3C protein (EBNA3C) degrades the retinoblastoma protein [27]. EBNA3C also compromises cell-cycle arrest by inhibiting the cell's p53 activity and fostering degradation of p53, thus negating p53 cooperation with the retinoblastoma protein to impose cell-cycle arrest [28–31]. EBV's latent membrane protein 1 (LMP1) is negatively associated with the presence of the retinoblastoma protein, suggesting an inhibitory effect of LMP1 on retinoblastoma expression [32]. EBV's latent membrane protein 2A (LMP2A) contributes to abrogation of cell-cycle arrest by methylating the PTEN promoter, thereby blocking PTEN's control of cyclin D, favouring the release of E2F from the retinoblastoma protein and, hence, the breaking of cell-cycle arrest [33,34].

With regard to cell survival, the negative effects of EBNA3 on p53 compromise apoptosis. LMP1 and LMP2A have anti-apoptotic effects through the NFkB pathway [35] at least in part by upregulating the anti-apoptoic protein survivin [36]. LMP2A enhances the anti-optotic effects of Bcl and Ras [37].

EBV also raises the limit on the total number of divisions a cell can undergo by relaxing control of telomerase reverse transcriptase (TERT), which maintains the length of telomeres during division. LMP1 and the Epstein Barr nuclear antigen 2 (EBNA2) contribute to this cell immortalization by inducing TERT expression and telomerase activation [38].

LMP1 abrogates the cell adhesion barrier to metastatic cancer [39,40]. Accordingly, its expression is robustly associated with metastasis in nasopharyngeal cancer [41].

EBV proteins also compromise restraints on oncogenesis in precise ways. LMP1, for example, promotes the expression of insulin-like growth factor 1 (IGF1) and phosphorylation of its receptor, thus favouring proliferation of infected cells [42]. Because IGF1 supports regulatory T cells [43], the promotion of IGF1 by LMP1 also probably reduces immunological destruction of EBV-infected cells.

Other DNA viruses compromise the same barriers to oncogenesis even when they have much smaller genomes. These viruses rely more heavily on the multiple actions of each viral protein. HPV, for example, has only about one-tenth the number of genes of the EBV genome. Oncogenic types of HPV abrogate barriers to oncogenesis primarily through multiple effects of their E6 and E7 proteins. To break cell-cycle arrest, E7 binds to retinoblastoma protein, causing E2F to be released from the retinoblastoma protein [44], thereby initiating DNA synthesis. E7 also binds to cyclin-dependent kinase inhibitor 1 directly, thus providing another mechanism for compromising cell-cycle arrest [45]. E6 binds to E6-associated protein (E6AP) to mark p53 for degradation, thus negating p53's participation in cell-cycle regulation and apoptosis [46,47]. E6 also contributes to cell-cycle arrest by binding to p150/sal2, which is a transcriptional activator of cyclin-dependent kinase inhibitor 1 [45].

The interaction between E6 and E6AP activates telomerase by marking for degradation an inhibitor of TERT (NFX1-91) [48,49]. E6 may also contribute to telomerase synthesis by hypomethylating the TERT promoter [50] and by associating with the activating variant of the NFX1 molecule (NFX1-123), thus enhancing hTERT expression [51].

The different tumour viruses illustrate variations on the theme. EBV, HPV and the other oncogenic viruses (table 1) illustrate how viral oncogenicity has evolved convergently [10], though sometimes incorporating distinct mechanisms; McPyV, for example, triggers cell division as a result of a mutation in its genome that recurs during the course of an infection [52]. Although the biochemical mechanisms differ, the same barriers are abrogated. The applicability of these generalizations to the spectrum of oncogenic viruses has been assessed in the context of the hallmarks of cancer [10] and the barrier theory [7].

The spectrum of accepted virally caused cancers has increased gradually since Luria's speculations on infectious causation. The extent to which more cancers will be recognized as having viral causes may depend on attention to additional cryptic aspects of viral oncogenesis. We suggest that two related aspects of crypticity may presently be responsible for a delay in acceptance of viral causation of cancers: a low ratio of viral to host genome number and inconsistencies in replicability among studies.

3. Interpreting low viral loads in tumours

EBV illustrates cryptic viral causation of cancers that are characterized by low and high viral loads. In the cancers for which EBV was accepted decades ago—endemic Burkitt's lymphoma, nasopharyngeal carcinoma, post-transplant lymphoma—the ratio of viral genome to host cell tends to be greater than one [53,54]. A low viral load has been used as evidence that EBV is not causing other cancers [54–57]. The logic underlying this assumption is that virtually all cells of a tumour are cancer cells and that the most conspicuous examples of viral causation can be used as a benchmark for rejecting viral causation if viral counts are low.

Several considerations suggest that oncogenic viruses can be primary causes of cancer even when they infect only a small proportion of the cells in a tumour. The most obvious consideration is the infiltration of the tumour by uninfected cells. Less conspicuous effects arise from selective destruction of infected cancer cells by cytotoxic T cells and a pro-proliferative microenvironment that may cause non-cancerous cells to proliferate in response to the activity of the cancer cells.

Asymmetric transfer of viral genomes during cellular division may also contribute to a low proportion of infected cells within a tumour. EBV genomes may persist episomally, but the episomes are not transferred reliably to daughter cells during cell division [58]. They therefore would decline in frequency in the tumours, a decline that is at least partly compensated for by enhanced proliferation of the infected cells [58].

If infected cells are being selectively destroyed by the action of cytotoxic T cells, the continual production of uninfected cells could cause the ratio of infected cells in such cancers to decline. For EBV, the importance of this destruction of infected cells is reflected in counter-adaptations of the virus [59]. Its EBNA1 protein, for example, reduces MHC-1 presentation of EBV antigens, enhancing viral persistence by reducing the destruction of infected cells by cytotoxic T cells [60].

4. Hodgkin's lymphoma and crypticity of viral causation

Hodgkin's lymphoma is a human cancer for which a causal role for EBV has been accepted in recent years [26,61–63]. This cancer is characterized by the presence of enlarged B lymphocytes, which are referred to as Hodgkin's and Reed/Sternberg cells (HRSCs) when they are uninucleate or multinucleate, respectively [61]. HRSCs are now accepted as the cancerous cells in Hodgkin's lymphomas even though they comprise only about 1% of the cells in the tumour [61]. These cells are virtually all infected, whereas other cells in the lymphoma are rarely infected [62]. This low frequency is an aspect of the crypticity of the EBV aetiology of Hodgkin's lymphoma. It has been explained at least in part by a high frequency of infiltrating cells [61]; however, the frequency of infected cells could also be lowered by the processes mentioned above: infiltrating cytotoxic T cells that selectively kill EBV-infected cells, asymmetric transmission of EBV episomes and a pro-proliferative microenvironment [59,64].

Although the HRSC phenotype is a marker of cancer cells in Hodgkin's lymphoma, it can occur in cells that are not cancerous. In infectious mononucleosis, EBV enhances proliferation of infected B cells, some of which have Hodgkin/Reed Sternberg morphology [61]. This finding suggests that EBV can generate the HRSC phenotype prior to the transition of infected B cells to a cancerous state. It is thought that EBV's LMP1 protein contributes to the HRSC phenotype [65].

Another aspect of crypticity is that EBV has been found in approximately 30–80% of Hodgkin lymphomas [53,61,62]. If a virus is found in virtually all cases of a particular cancer, as HPV is in cervical cancer and EBV is in endemic Burkitt's lymphoma, this uniform presence reduces the resistance to accepting an etiological role for the virus. If a virus is found in a small portion of patients its causal association seems less compelling. In the case of Hodgkin's lymphoma, reports of a stronger association (close to 80% of study populations) has helped garner acceptance of EBV causation, but the spread of positivity raises concerns about the research techniques on false positivity or false negativity, which could be exacerbated by low viral loads. Alternatively, the range of positivity among patients could reflect real differences in EBV causation among study populations. This possibility mirrors that of hepatocellular carcinoma, which can be caused by HBV or HCV (table 1).

Overall, Hodgkin's lymphoma illustrates how crypticity of viral causation may contribute to a relatively recent acceptance of infectious causation. These aspects of inconspicuousness may help inform analyses of cancers that have been associated with viruses but for which viral causation has not yet been accepted.

5. Human cytomegalovirus and glioblastoma

A link between the aggressive brain tumour glioblastoma and the DNA virus human herpes virus 4 (human cytomegalovirus, HCMV) has been proposed with different research laboratories reporting evidence of HCMV in tumour samples (e.g. [66–68]) and others finding limited or no indication of HCMV presence (e.g. [69]). Despite lack of consensus regarding the role of HCMV in this brain cancer, treatment targeting the virus (antiviral therapy) or its peptides has shown promising results in prolonging survival [70,71]. Although HCMV is presently considered to be an ‘oncomodulatory’ rather than an oncogenic virus, it can transform cells in vitro, abrogate the four major barriers to oncogenesis and contribute to each of the hallmarks of cancer [67]. Preference for the term oncomodulatory has been based on differences from accepted oncogenic viruses such as lack of both integration into the cellular genome and sustained expression of oncoproteins and the fact that it has not yet been accepted as a cause of any human cancer [67]. The ability of HCMV to abrogate the four major barriers to oncogenesis, however, indicates that it has the critical oncogenic attributes specified by the barrier theory of oncogenesis.

The intricate evolutionary arms race between HCMV and the host may make it difficult to determine the role of HCMV in oncogenesis. While HCMV is considered an expert at immune escape, it is important to note that the majority of humans are infected and have few if any associated symptoms [72]. However, infections in individuals with nascent or compromised immune systems reveal that HCMV can be a damaging and deadly virus. HCMV infection is life-long and for most individuals is kept in check by the immune system. HCMV persists in a non-virion producing state (from which it can reactivate), using its wide array of viral gene products to push the host cell to proliferate indefinitely and evade apoptosis [73–75].

Other sophisticated efforts to avoid detection by the immune system are exemplified through HCMV's ability to interfere with viral antigen presentation by the major histocompatibility complex (MHC) proteins on both antigen presenting cells (MHC class II) and in the remaining wide variety of human cells that HCMV can infect (MHC class I). HCMV encoded proteins are involved in hiding viral antigens from MHC recognition and degrading or sequestering MHC molecules [76]. Cells altogether lacking MHC I presentation may be detected and destroyed by natural killer (NK) cells. As part of its repertoire for NK cell evasion, HCMV encodes an MHC mimic that traffics to the cell surface to subvert NK destruction [77].

Despite its immune evasion toolbox, evidence of a strong immune response is demonstrated by a diverse T-cell population capable of recognizing multiple HCMV antigens in healthy hosts. In older adults infected with the virus, T cells that specifically recognize HCMV can make up as much as half of the memory T-cell population [76]. Macrophages, which are able to present antigen and activate T cells, appear to avoid MHC manipulation by HCMV and may play a major role in initiating and sustaining the T cells critical to containing infection [78]. The immune system's ability to circumvent viral evasion means that identification and destruction of HCMV-infected cells should be ongoing.

Glioblastoma is characterized by within tumour heterogeneity [79] and the presence of cancer stem cells [80]. Investigators have found a ratio of HCMV to tumour cells of approximately 0.1 [81]. In vitro experiments indicate that infecting glioblastoma cells with HCMV results in a stem cell-like phenotype [82,83], possibly suggesting that a limited number of infected cells could be driving tumorogenesis. Virions have not been detected in glioblastoma tissue, thus HCMV proteins found in these tumours may be involved in long-term viral survival rather than viral production [67]. Using glioblastoma as a model, the robust host immune vigilance and response to HCMV coupled with HCMV's multifaceted ability to dysregulate cellular behaviour such as cell-cycle arrest and apoptosis when semi-latent, may result in an oncogenic effector that is hard to detect in cancer tissue.

6. Breast cancer as an illustration of potential crypticity

When a particular cancer is caused by many viruses, the oncogenic role of each may be difficult to ascertain. The following viruses have been correlated with breast cancer: mouse mammary tumour virus (MMTV, also referred to as mouse mammary tumour-like virus or human mammary tumour virus to designate isolation from humans), EBV, HPV, HCMV and bovine leukaemia virus (BLV). For each virus, the presence and absence of associations have been reported generally by multiple research teams [84–90]. (MMTV and BLV are retroviruses, but control cellular activity through an integrated DNA version of their genome, and therefore act like DNA viruses, within the cell, as does HTLV-1, the only human retrovirus that has been an accepted as a direct cause of human cancer; table 1.) For each of these viruses, studies that report associations find an elevated positivity in breast cancer samples (or breast cancer patients) relative to control tissues (or control patients) of roughly 20–50%. If each virus caused breast cancer independently, all five together could therefore account for all human breast cancer. Any contribution of these viruses to breast cancer, however, might vary greatly among populations (e.g. [91] for MMTV and [92] for BLV) and may act in concert rather than separately [5,93]. The extent to which different viruses may play a role in a particular category of cancer is just beginning to be investigated. The recognition that more than one virus can be contributing to a given category of cancer (e.g. HBV and hepatitis C virus for hepatocellular cancer and human immunodeficiency virus and HPV for cervical cancer) draws attention to the need to investigate the possibility that two or more virus may contribute to particular cancers.

Infiltrating cells may also contribute to the crypticity of viral causation of breast cancer. Infiltrating cytotoxic T cells are positively associated with overall survival and disease-free survival in two major subtypes of breast cancer (triple negative and HER2 positive) in both early stage and advanced disease [94–100]. This association suggests that cytotoxic T cells are effective, albeit imperfectly, at destroying breast cancer cells; it therefore raises the caveat that the low frequency of virus to host cell genomes is not by itself compelling evidence against a causal role for these viruses in breast cancer. An important next step would be to determine whether tumour-associated cytotoxic T cells are specific for viral or host cell antigens.

Microdissections of breast cancer show clusters of EBV-infected epithelial cells in the midst of large numbers of uninfected cells [55]. This pattern would be expected if infected cancer cells were able to multiply and disperse prior to detection and destruction by T cells. The infection of epithelial cells also argues against the hypothesis that the presence of EBV in the breast tumours results solely from the infiltrating, EBV-infected B cells, as does the statistical accounting for B-cell inifltration (e.g. [101]). Low virus-to-cell ratios in breast cancer have been compared with low levels in adjacent tissue. Similarity in these ratios has led to the conclusion that the virus is not causally involved (e.g. [54]). Although these comparisons are appropriate and can be informative, they also come with a caveat. Similar low positivity could occur even if the virus is causing the cancer, because immunological activity against infected cells could be greater in the inflammatory microenvironment of a tumour than in surrounding tissue. Such a similarity has been demonstrated for HBV positivity in hepatocellular cancer [56].

These considerations suggest that breast cancer may be a paradigmatic illustration for the potential importance of attentiveness to crypticity of infectious causation of cancer. Each candidate virus compromises most if not all of the major barriers to oncogenesis; however, associations with breast cancer that are confirmed by many studies and usually by several different investigators are not confirmed by other investigators. The large number of viruses associated with breast cancer suggests that the focus of investigators may generate different associations. Breast tissue is prone to cellular infiltration, which may dilute the virus signal, and cellular infiltration by cytotoxic T cells may selectively reduce the abundance of infected cells. Finally, two of the candidate viruses, EBV and HCMV, infect as episomes and may thus be lost from a substantial portion of the tumour cells.

7. Virally induced stemness and the stem cell theory

The hypothesis that infected cells could comprise a small portion of the cells in a tumour and still be the only cancerous cells in a tumour bears some similarities to the stem cell theory of cancer, which proposes that only a portion of the cells have stem-cell like characteristics. Primary lines of evidence advanced in support of the stem cell theory include the tendency for only a small portion of the cells in a tumour to be able to regenerate the cancer in animal models and the tendency for metastases to regenerate cells that are similar to those in the primary tumour [102]. These findings are consistent with the idea that tumours can be driven by a small portion of infected cancerous cells, which we refer to as the infected stem hypothesis, because the viruses confer stem-like characteristics on cells and associations of viruses with uninfected cells could yield the heterogeneity found in both primary and metastatic tumours. The heterogenous cellular composition could be derived from infected cancerous cells that have lost the infection, as is suggested above for EBV. Or they could result from effects of infected cells on nearby non-cancerous cells (e.g. through cytokine effects) or infiltration. The infected stem hypothesis proposes that the uninfected tumour cells, though proliferative in the microenvironment of the infected stem-like cells, are not actually cancer cells. Some investigators of tumour viruses have been moving in this direction. HPV, for example, has been proposed as the initiator of stem cell oncogenesis in oropharyngeal cancer [103].

One line of testing could compare the transplant potential of cancers with different virus-to-cell ratios. The stem cell hypothesis predicts that success at the generation of new tumours will correlate with the virus-to-cell ratios of the source tumours.

8. Implications for modelling of viral oncogenesis

A central issue in the study of cancer is the extent to which viruses cause cancer. Infection with an oncogenic virus can compromise all four of the major barriers to cancer simultaneously. For oncogenesis to occur strictly by mutations, a series of mutations would have to compromise these barriers. They would have to occur in sequence before the occurrence of mutations or other factors make the cell lineage non-competitive. If a mutation compromised cell-cycle arrest, for example, mutations that abrogate regulation of telomerase would have to occur before the limit on cell division imposed by loss of telomeres is reached. The probability of acquiring such a series of mutations without making the cell unable to outcompete normal cells must therefore be very low relative to the probability that all these barriers can be compromised at the onset of an infection with an oncogenic virus (which is essentially a probability of 1.0).

This discrepancy has been almost entirely ignored in considerations of oncogenesis, probably because the causal role of mutations without infection was accepted prior to the recognition that viruses can mastermind the precancerous phase of oncogenesis. Mathematical models comparing the probabilities of generating cancer by infection and mutation relative to mutation alone have not yet been published. They could provide important insights for comparing the feasibility of mutations without infection as an oncogenic process relative to mutations with infection.

Another critical area for model development involves evaluation of the net effects of the generation and infiltration of uninfected non-cancerous cells in a tumour together with the differential reproduction and death of infected cancer cells relative to non-cancerous cells. Agent-based models that assess the relative proportions of infected and susceptible cells in the context of selective immune destruction of infected cells seem particularly applicable. Agent-based modelling of breast cancer, for example, has not incorporated viral interactions [104]. Inclusion of viruses into such models might offer a resolution to the ambiguities pertaining to viral load if they encompass the spectrum of possibilities concerning infiltration of virus-specific cytotoxic T cells, loss of episomes and pro-proliferative environmental effects.

These opportunities for generating insight into the feasibility of infectious causation of cancers may be centrally important for the goal of preventing and curing cancer. The current state of knowledge drawn together in this paper indicates that viruses may be causing cancers more inconspicuously than has been presumed as a result of low viral loads in the tumour. If theoretical analyses indicate that oncogenesis without infection is much less probable than infection-dependent oncogenesis and that a low frequency of virally infected cancer cells is a likely result of the interaction between cancer cells and immunological defences, then much of the cancer of uncertain cause may turn out to be caused cryptically by infection. One of the strengths of health sciences is the prevention and control of infectious diseases. The first step in advancing this process is identifying an infectious cause that can be targeted.

Data accessibility

This article has no additional data.

Competing interests

We declare we have no competing interests.

Funding

Development of the ideas in this paper were supported by grants from the Rena Shulsky Foundation and The Breast Cancer Fund at the National Philanthropic Trust at the recommendation of the National Breast Cancer Coalition awarded to P.W.E.