Introduction
Polly Matzinger’s first published danger model of immunity ( [1] ) proposed only one mechanism for immune recognition of danger—the dispersal of cellular contents following necrosis of a diseased cell. Since then many specific molecular signals have been shown to alert the immune system to the presence of danger in the form of infection. Pathogens often bear evolutionarily conserved molecules not present in healthy hosts—pathogen associated molecular patterns (PAMPs) ( [2] ) such as lipopolysacharides, for which we have evolved a set of pattern recognition receptors. In addition, host cells often alert the immune system to their infection with molecular signals including MICA and ULBP (2). The danger-recognition, intercellular communication pathways so far proposed rely on antigen presenting cells (APCs), most potently dendritic cells, to absorb and process antigens (Ags) for presentation to T cells, and also to recognize associated danger signals.
T cells, too, measure danger signals on host cells they contact. Each clonal cohort of T cells is specific to one Ag—its cognate Ag. Since the immune system as a whole must correlate Ags to disease, it makes great sense that T cells should gather danger information, both because of their Ag-specificity and because they continually contact host cells throughout most tissues in the body as they scan them. A T-cell-mediated danger-recognition pathway would provide the immune system with several advantages not always available to pathways relying on dendritic cells and other APCs, including:
- A direct pathway for the recognition of danger from non-lytic viruses which can lead to rapid commencement of killing of infected cells.
- Relatively rapid deactivation of activated T cells whose cognate Ag is not correlated to danger, even during an infection when danger signals abound.
- An alternative pathway, not dependent on APCs, so that the immune system can function even if its APC compartments are compromised.
The danger of non-lytic viruses
Non-lytic viruses do not kill the cells they infect; instead of bursting the cell open to escape, the viral progeny bud out from the cell membrane. Consequently the Ags of non-lytic viruses do not get collected by the same APC mechanisms that collect Ags from necrotic host cells. Matzinger has written “There is also no need to make a response to a virus that enters a cell, makes few copies of itself and then leaves without doing any damage. (We might even want to welcome such viruses for the genes that they could bring us.)” ( [3] ).
However, even non-lytic viruses do pose real danger to the host. At best, they consume cellular resources, and their proteins and genetic instructions can cause random interference with proper cellular function. At worst, evolved viral interference with cellular growth signals and apoptotic (cellular suicide) mechanisms can cause malignancy. Studies have shown that non-lytic as well as lytic viruses generate immune reactions, although the cytopathology of the virus can influence the type of immune response ( [4] , [5] ). Non-lytic viruses appear more likely to trigger cytotoxic responses from killer T cells than necrotically lytic viruses which tend to trigger antibody mediated defense [4].
T cells kill virally-silenced bystanders with FasL
Fas ligand (FasL) has been shown to induce apoptotic cell death when it binds to its receptor, Fas, on host cells. However target cells often are not killed by Fas:FasL interactions since cells have widely ranging Fas sensitivities. Such different responses can be explained in part by differences in the quantity of Fas receptors constitutively expressed, in the ability of targets to upregulate Fas, and in the states of both pro- and anti-apoptotic pathways.
One of the factors determining a cell’s Fas sensitivity is viral infection ( [6] , [7] ). FasL has even been proposed as a specifically antiviral killing mechanism used by T cells ( [8] ). Therefore, the response of a cell to challenge by FasL could provide a clue as to whether it had been infected. An immune system that makes the fullest use of danger information best protects its host, and so is strongly favored by evolution. Therefore, one may reasonably hypothesize that some immune systems have evolved to exploit sensitivity to FasL, as a means of discovering danger.
Smyth and coworkers showed that after T cells kill targets displaying the T cell’s specific Ag, the T cells then kill FasL-sensitive bystanders using only FasL ( [9] ). Smyth et al. suggested that bystander lysis may be necessary because many local cells may be infected even though they do not display viral proteins, since viruses are known to negatively influence Ag presentation through a variety of mechanisms ( [10] , [11] ).
Do T cells discover aberration with FasL?
Major histocompatibility (MHC) molecules display Ags, typically fragments of proteins, on the cell surface. A T cell receptor (TCR) on a T cell, binds the complex comprising an Ag and the MHC molecule that presents the Ag on the exterior of a host cell. The TCR’s chemical structure accounts for the selectivity of T cells; each TCR is maximally responsive to one Ag, the T cell’s cognate Ag, or a very small number of Ags. The MHC molecules of mice are called H‑2 molecules. Laboratory mice have been inbred into immunologically identical strains whose MHC types are indicated by various suffixes, including b, d, and k.
Smyth has also published results of cytotoxicity assays against various direct and bystander targets using lymph node populations harvested after repeatedly vaccinating mice against splenocytes from immunologically disparate (allogeneic) mice ( [12] ). In most assays, the T cells (H-2b) killed allogeneic cells (H-2k). The T cells, from a vaccinated host’s lymph nodes, thus saw the primary target cells (H-2k) as being foreign and similar to the cells used to vaccinate the T cells’ host. Many assays included syngeneic (H-2b, identical to the T cells) bystander targets as well as the direct targets (H-2k).
Some assays compared the number of bystander cells killed in conjunction with T cell lysis of either L929 or L929-Fas cells (L929 cells stably transfected with human Fas cDNA) as primary targets. Smyth’s Fig. 3B showed two to three times more killing of syngeneic bystanders in conjunction with lysis of L929-Fas primary targets, than in conjunction with lysis of Fas-negative L929. Smyth noted the effect in his text, in passing. Qualitatively similar effects are also shown in Smyth’s Figs. 4A and 4B.
The T cell danger-correlation hypothesis interprets the intercellular interactions depicted by Smyth as follows. T cells first find their primary targets by searching for cells displaying the T cell’s cognate Ag. After delivering lethal hits to their primary targets, using both perforin and FasL, the T cells then challenge bystander cells with only FasL. The extent to which T cells challenge bystanders is proportional to the degree of activation the T cells retain after attacking their primary targets. The hypothesis further posits that T cells decrease their activation level if they discover while attacking their primary targets, that those targets do not yield signs of disease. Indications of disease may include presentation of Fas receptors. The results, showing decreased lysis of bystanders in conjunction with Fas-deficiency of primary targets, would be consistent with this explanation, and otherwise anomalous. The bystander cells are challenged with only FasL, not perforin which would kill all bystander targets, in part because a function of the T cell interaction is to measure the aberration of the bystander cells for comparison to cognate targets.
Danger receptors on T cells
FasL deployed on a killer T cell not only delivers a pro-apoptotic message to virally infected target cells, but signals the T cell to proliferate when engaging Fas receptors as well as cognate Ag on targets ( [13] , [14] ). This result is consistent with the premise that T cells react to the presence of Fas on target cells as if it were a danger signal. FasL is just one member of the Tumor Necrosis Factor (TNF) family of ligands. Another TNF ligand that induces apoptosis in target cells, especially virally infected ones ( [15] ), TRAIL (TNF-Related Apoptosis-Inducing Ligand), also sends a stimulatory signal back to the T cells that deploy it ( [16] ). Therefore T cells have access to multiple pathways to sense aberration in target cells while challenging them with potentially apoptosis-inducing ligands.
When MIC, which is typically displayed by infected cells, ligates the NKG2D receptor of T cells at the same time the T cells detect their cognate Ag, the T cells are stimulated to proliferate and release IL-2 ( [17] ), a potent cytokine which directs other immune cells to target infected cells. Thus tissue cells can costimulate T cells through the T cells’ NKG2D receptors which substitute for their CD28 receptors (17), perhaps even providing signal 2. Such costimulation provides an activation mechanism alternative to that of Matzinger’s original danger model in which “non-APCs (non-hematopoietic tissues) cannot provide signal 2, and so tolerize resting naïve and experienced T cells ( [18] )”.
Given the similarities in the responses of T cells to ligation of their NKG2D or CD28 receptors, danger signals on living tissue cells might even provide a mechanism for priming naïve T cells. Such a mechanism would be an alternative to the classical priming mechanism relying on APCs. It would cause autoimmune disease unless the priming sequence required that healthy reference cells be measured to be free of both danger signals and the relevant antigen. However, it would speed the detection of, and improve the defense against, viruses that do not burst host cells and release large quantities of molecules indicating pathology amid viral fragments and whole viral particles—a combination which identifies viral proteins to the immune system through the APC-based priming mechanism.
A self-referential danger model
Infected cells often recruit the immune system to kill them as an alternative to committing suicide (undergoing apoptosis) which is a common response of host cells to becoming diseased. Medzhitov and Janeway have noted that one possible reason for the recruitment of immune killer cells in lieu of simple apoptosis, is that recruiting the immune cells can also trigger them “to produce cytokines such as interferon γ” (2) which will lead to identification and killing of other infected cells in the vicinity. The danger-correlation hypothesis proposes that diseased host-cells recruit the immune system’s assistance in their death also so that their disease-specific Ags can be identified, not just to sound a general alarm that an infection is present. Even if T cells do not use healthy self cells as a comparative reference, the mere fact that MIC ligation of NKG2D induces proliferation of only those T cells contacting cognate Ags (17) implies that T cell clones specific to the pathogen, will undergo greater proliferation.
Janeway has further asserted that “the adaptive immune system of T lymphocytes and B lymphocytes is always referential to self, as it is selected on self-ligands; it persists in the periphery on self-ligands; and at least for T cells, it is dependent on self-ligands to be able to mount a response” ( [19] ). Experiments by several groups have shown that killer T cells die or fail to proliferate in the periphery without contact ( [20] ) with self-peptides ( [21] ) presented on MHC molecules, yet can proliferate moderately in vivo when presented with self-peptides, even without contact with their cognate (typically foreign) Ags. Such experiments usually transplanted T cells to syngeneic recipients whose own immune cells had been depleted, in which case the transplanted T cells expanded in numbers more than in recipients with normal quantities of immune cells. The required self-reference for T cell proliferation has generally been interpreted as being a mechanism of maintaining normal cell counts.
In contrast, the danger-correlation hypothesis proposes that the self-reference T cells require, allows them to quantitatively compare danger signals on cognate (typically infected) vs. reference (typically healthy self) targets. T cells that cannot make such a comparison are of relatively low-utility, perhaps even harmful, to the host, so evolution has directed them to undergo apoptosis when they cannot detect self-reference targets.
Why reference targets should be syngeneic
Smyth and coworkers showed that to be challenged after attacks upon primary targets, bystander cells must be syngeneic with the T cells. “B6.P0 anti-E749-57 CTL [Cytotoxic T Lymphocytes] that were activated with RMA-E7 cells lysed labeled syngeneic bystander [ H-2b] B6 blasts but did not significantly lyse H-2k C3H/HeJ blasts, H-2d BALB/c blasts, B6.β2µ0 blasts [which are syngeneic with the T cells, but lack a key component of their MHC molecules which may preclude the T cells from recognizing their syngeny], or FasL-insensitive [H-2 b] EL4 cells... preliminary data (not shown) also suggest that alloreactive mouse CTL (b anti-k) do not effectively lyse H-2 d BALB/c blasts acting as bystanders. Further experiments with B6 anti-E749-57 CTL were performed, and their mode of bystander lysis was identical to that of B6.P 0 anti-E749-57 CTL. In particular, perforin-mediated bystander lysis of allogeneic blasts was not observed, even after pretreatment of B6 anti-E749-57 CTL with PMA-ION...” (9)
Their data confirm an earlier study by Duke ( [22] ), although Kojima et al. ( [23] ) have presented evidence of bystander lysis of non-syngeneic bystander L1210-Fas cells derived from a lymphocytic leukemia cell line. Perhaps the leukemia cell line, unlike the non-malignant blast cells Smyth used as bystanders, displays enough danger signals in addition to Fas, so that T cells attack it if found in the vicinity of their cognate targets. In that case the bystanders were not being used as reference targets but were challenged with FasL by the T cells’ mechanism that kills cells infected by viruses that have silenced MHC display of viral Ags.
The danger-correlation hypothesis proposes that T cells find syngeneic, non-cognate (bystander) targets to measure as a reference to compare to their cognate targets. By challenging only such presumably healthy bystanders with FasL, the T cells can then compare danger signals such as Fas presentation, of average cells, to those of their cognate targets. Different tissues in the body vary in the amount of Fas receptors they display and their sensitivity to ligation of those receptors; if T cells did not use reference measurements of Fas display, they would be prone to attack healthy cells in Fas-rich tissues, and might fail to kill infected cells in low-Fas tissues.
Why must non-aberrant bystanders be syngeneic with the T cells in order to be challenged with FasL? A T cell must engage the MHC molecules displayed on a reference-target to verify it does not display cognate Ags, thus part of the reason for the finding that only syngeneic bystanders are challenged.
More important, if T cells merely found the MHC molecules of reference cells without verifying their syngeny using a polymorphic (varying between individuals) locus of the MHC, then the T cells would be vulnerable to viral ruses. For example, after cytomegalovirus gains control of a host cell, it prevents display of the host’s MHC molecules and replaces them with its own homologues that may not effectively display viral peptides (11, [24] ). If the virus then increased the level of danger signals such manipulated host cells display, it would decrease the amount of immune attack upon infected host cells that had not yet been manipulated to hide viral peptides. The immune system would be tricked into using the fully compromised, infected cells as a danger reference, and would measure a negative correlation of danger to viral peptides.
By checking the MHC-type of the reference cells, T cells limit the ability of virally infected cells to appear suitable for use as reference cells for danger measurements. Although a virus could evolve to mimic any single MHC allele and deceive a single host, it would not be able to use that deception on a new host possessing different MHC types.
[This paragraph was added after original publication:] In contrast, the conventional explanation for cytoxicity directed against bystanders, is that since they might be infected with a pathogen that compromises their Ag presentation, they must be killed as a precaution. However, the experimentally demonstrated requirement for syngeny would be counter productive in this context; it would be even easier for viruses to completely eliminate Ag display, and thus abrogate the sygeneic signal, than it would be to silence only the display of viral peptides. The sygeneic signal presumably is provided by display on MHC, of an endogenous peptide, probably very similar in peptide-sequence to the T cell's cognate peptide.
Superantigens and killer T cells
Some pathogens secrete superantigens which affect the T cell receptors (TCRs) of a substantial percentage of T cells regardless of their cognate Ags. The immediate effect of stimulating many T cells would be to increase their destruction of infected cells as well as causing systemic inflammation, thereby reducing the host’s viability and ability to inadvertently spread the pathogen to other hosts. What evolutionary advantage do pathogens gain from activating large numbers of T cells? Fuchs and Matzinger have shown that such activation of virgin helper T cells can tolerize them as they encounter B cells while their TCRs are still ligated by the superantigens ( [25] ). But what of killer T cells, which are also sometimes the targets of superantigens?
Superantigens, bound to a T cell’s TCRs, would prevent it from distinguishing between targets bearing cognate Ags, and reference targets not bearing cognate Ags. The thresholds of sensitivity a T cell has to cognate Ag display, would critically affect its response to superantigens. The optimum behavior of a T cell would probably be to kill cognate targets only when they display cognate Ags above a relatively high threshold of concentration, and to use cells as danger references only when they display cognate Ags at concentrations below a much lower concentration level. If that is the case, then high concentrations of superantigens would prevent T cells from distinguishing between cognate targets which they should kill, and normal host cells. Thus high concentrations of superantigens would render the T cell useless at best, and dangerous to the host at worst. Lower concentrations of superantigens would prevent the T cells from recognizing non-cognate cells for danger references. Thus lower concentrations of superantigens would prevent T cells from correlating danger signals to antigens.
Perhaps T cells die or become anergic after being unable to detect non-cognate cells after killing a certain number of cognate targets. Such behavior would assure that only T cells actively and continuously correlating their cognate Ag to danger signals, could indefinitely continue to kill host cells. Such behavior could also prevent massive cytotoxicity to healthy host cells if T cells gradually accumulated superantigen on their TCRs. Relatively low concentration of superantigen ligation of TCRs would prevent the recognition of non-cognate reference targets; as long as the accumulation of superantigen was not so fast that the higher threshold of cognate target recognition was reached before the T cells ceased cytotoxic activity, healthy host cells would be protected. This proposed limitation on killing by T cells could explain how superantigens effect mass tolerance among killer as well as helper T cell populations, and why superantigens more typically cause tolerance than toxic shock .
Dendritic cells and early viral proteins
Many viruses interfere with the display of Ags on the MHC of host cells. Their viral proteins that are expressed relatively late in the colonization of a host cell, may virtually never be displayed on the cell’s MHC molecules. Only the early proteins of such viruses might ever be externally observable on host cells during infection, and perhaps only before the cell has detected the infection and responded by displaying danger signals. Such viruses would pose great problems for an immune system that relied on patrolling Ag-specific cells to correlate danger signals to protein fragments.
The host could detect such viruses by embedding stationary, non-Ag-specific, Ag-collecting cells in various tissues. Such cells would collect Ags secreted by, or displayed on the surface of regular tissue cells, and would store them quiescently for a brief time period. If a tissue cell first displayed or secreted some Ag, and then later displayed danger signals indicating possible infection, the Ag-collecting cell could then present the stored Ags to patrolling T cells along with costimulatory molecules which would so highly stimulate the T cells that they would attack host cells displaying those Ags for some time, even in the absence of a correlation between those Ags and danger signals.
Such stationary, Ag-collecting cells could not afford to be Ag-specific because the number of possible, distinguishable Ags is much greater than a billion; there could not be enough Ag-specific cells embedded in tissues to catch most infection-related Ags quickly wherever infection first strikes. However, the combination of embedded, stationary, non-Ag-specific, Ag-collecting cells; and patrolling, Ag-specific T cells that are activated by the Ag-collecting cells, would suffice. The efficiency of such a system would be enhanced if the embedded, Ag-collecting cells had dendrites which stretched among the tissue cells allowing one Ag-collecting cell to monitor many tissue cells. Such an Ag-collecting cell could constitute another kind of comparative, self-referential, danger-monitoring system; however, instead of comparing the danger signals between cells bearing or not bearing specific Ags, it would measure how danger signals from individual tissue cells change with time or vary between initially similar tissue cells.
Any such quantitative, comparative self-reference would allow the immune system to accommodate variations in danger signals between tissue types, and variations between individuals in predisposition to present danger signals. Genetic diversity in immune response is critical to protecting a species even when immune protection fails individuals; comparative, self-referencing danger-sensing schemes can play a crucial role in allowing diversity in danger signaling.
The embedded, Ag-collecting cell thus hypothesized, has the known behavior and morphology of dendritic cells. The known great potency of dendritic cells to activate T cells therefore follows as a likely consequence of the hypothesis that T cells otherwise become tolerant to Ags in the absence of relatively high danger signals presented by cognate targets at least sometime in the infection cycle.
Dendritic cells absorb Ags that are bound to heat-shock protein 70 and secreted by tissue cells, then present the Ags to killer (CD8+, via MHC class I) T cells to activate them against cells displaying those Ags ( [26] ). Circulating dendritic cells possess ILT3 (LIR-5) receptors which inhibit immune responses against cells presenting MHC molecules, and which also can internalize antigens for presentation to T cells ( [27] ). Such receptors, if also present in embedded dendritic cells, might allow them to respond to the disappearance of MHC molecules from tissue cells they contact, by presenting previously acquired peptides, such as fragments of early viral proteins, and costimulation, to T cells.
Peripheral tolerance
Most nascent T cells whose cognate Ags are common self-peptides, are deleted in the thymus. However, since not all self-proteins appear in the thymus, T cell tolerization must also occur in part in the peripheral circulation, but the mechanisms have remained unclear.
A danger model in which T cells continually correlate danger signals to their cognate Ags, predicts that T cells respond to low danger-correlation by becoming tolerant of their cognate Ag. A danger-correlation signal gradually developed inside T cells, could explain why tumor-targeted, adoptively transferred T cells become tolerant of their tumor-associated Ags (TAAs) after spending enough time in a new host with advanced tumors, even with a blockade of known tolerizing receptors ( [28] ).
Implications for cancer immunotherapy
By assuming that T cells probe for danger signals displayed by aberrant cells, an extended danger model would predict that the immune system recognizes many nascent tumors while they still display Fas and other danger signals. However, long-term interaction between tumors, which are genetically unstable and therefore evolve quickly to adapt to selective cell killing, and the immune system, tends to reduce the immunogenicity of the tumor ( [29] ).
One immune-evasion mechanism tumors develop as the immune system selectively kills the more immunogenic cells, is disruption of the Ag-presentation system. Immunoediting of tumors would likely also downregulate the display of danger signals by surviving tumor cells, and thereby induce peripheral tolerance to TAAs. Such induced tolerance could significantly augment the many previously proposed mechanisms preventing various tumor vaccines, that protect against tumor cell challenges after vaccination, from curing animals of tumors established prior to vaccination. A large established tumor, containing thousands of times more cells than the vaccination, exerts too powerful a tolerizing effect. Matzinger’s original danger model (1) specified that large tissue masses that did not signal danger, would gradually tolerize T cells to the tissue’s Ags. Such an effect could be hastened if T cells are able to find reference targets whose expression of danger signals, while typical of healthy tissue in that individual, is equal to or greater than those expressed by the tumor.
Reference cells and tumor tolerance
Groh and coworkers showed that for target cells with low-to-moderate expression levels of cognate Ag, T cells exhibit reduced cytotoxicity against targets with reduced expression of either MIC (which ligates NKG2D and thus is a danger signal) or the cognate Ag of the T cells. MIC alone was insufficient to trigger cytotoxicity, and very low concentrations of cognate Ag without MIC generated extremely small amounts of cytotoxicity (17) . This result is important not just because it hints at a correlation, between danger and antigen, taking place inside T cells, but also because it shows that early tumor cells have two divergent evolutionary paths to evade immunosurveilance.
(From Groh and coworkers's Costimulation... [17])
Figure 4. Antigen dose-dependent augmentation
of cytolytic T cell function by NKG2D.
Cytotoxic responses of pp65-specific T cells against C1R-A2-MICA
double transfectants pulsed with the HLA-A2–restricted NLVPMVATV peptide
were substantially stronger than those against identically treated C1R-A2
transfectants within a range of suboptimal peptide concentrations.
These increases were diminished by mAb to MICA or NKG2D.
The results obtained with the 4H6-254 T cell clone were representative
of five T cell clones tested. All assays were done in triplicate with deviations
that were not greater than about 3%. (Used by permission.)
A tumor is likely to begin with a population of cells that display TAAs on their MHC molecules, and also display danger signals due to their transformation into a malignant state. If some tumor cells evolve an elimination of TAA loading of their MHC, but still display MHC and many peptides typical of healthy self, they may serve as reference cells. They may escape killing by the host’s immune system since they do not also present cognate Ags to tumor-specific T cells, even though they might still present a high level of some danger-signals.
Another population of tumor cells may evolve to evade immunosurveilance by suppressing its display of danger signals, while still displaying cognate Ag. Therefore, in a possible resulting, heterogeneous tumor, T cells would encounter apparent reference cells with high levels of danger signals, and TAA-presenting cells with lower levels of danger signals. That combination should potently induce tolerance to TAAs since TAAs would appear to have a negative correlation to danger. The ability of such a heterogeneous tumor to evade immunosurveilance is one possible explanation why invasive tumors sometimes have increased expression of MIC ( [30] ).
Groh and coworkers have found that many tumors heterogeneously express MIC “with about 40–70% of cells showing intermediate to high levels of expression of MICA/B.” Many of the FACS outputs they showed as typical for MIC+ tumors were bimodal, meaning having two distinct populations of cells, one with heavy MIC expression, the other with notably less or no MIC expression. The expression levels graphed for ovary and colon tumors were very bimodal, showing one distinct population of tumor cells expressing undetectable levels of MICA, and the other population averaging noticeably more MICA than the MICA+ lung and prostate tumors graphed. As a whole, ovary and colon tumors were more likely to express MIC (5 of 6 and 4 of 5 tumors respectively) than lung and prostate tumors (2 of 6 and 2 of 4). Therefore the kinds of tumors (by organ source) that were more likely to express MIC, seemed more likely to have strongly bimodal populations, with the MICAhi population expressing more MICA than other kinds of tumors, and the MICAlo population expressing much less MICA (relative to detected control Ig) than tumor types less likely to express MIC ( [31] ).
(From Groh and coworkers's Broad tumor-associated expression...[31] Fig. 1. Surface expression of MICA on ... freshly isolated tumor cells. Tumor cell suspensions prepared from lung (B and C), prostate (D and E), ovary (F), and colon carcinomas (G) were stained with mAb 6D4 (solid profiles) or with isotype-matched Ig (open profiles) and examined by flow cytometry. B and D show examples of tumor cell preparations that were negative for binding of mAb 6D4 whereas C and E-G show positive stainings. These data are representative of the range of staining intensities seen with all of the tumor samples tested. Similar results were obtained with mAb 2C10. (Used by implied permission of the copyright holder and explicit permission of an author; PNAS makes this entire paper available via the internet for free.)
Perhaps the ovary and colon tumors measured were able to evolve evasion of immunosurveilance by bifurcating their cell populations into reference cells which did not express TAAs but expressed high levels of danger signals, and TAA-expressing cells that expressed low levels of danger signals.
An alternative explanation why late-stage or invasive tumors might abundantly express MIC, is that when they release soluble MIC, it binds to the NKG2D receptors on T cells, blinding them to the distinction between dangerous and non-dangerous cells ( [32] ). Such blinding is analogous to the way superantigens affect T cells. The only safe behavior for a blinded cytotoxic cell, is to refrain from killing any target, so evolution has directed lymphocytes to cease killing activity when they cannot distinguish dangerous cells from healthy cells. Requiring a positive correlation between cognate Ag and danger signals, to be continually detected in order for killing to continue, would force cytotoxic cells to behave in such a safe manner.
If reference target cells are not available, a condition more plausible in the center of a large non-heterogeneous solid tumor than in tissue infected by a haphazardly propagating virus, a likely immune response could be that cytotoxic cells would migrate out of the tumor or otherwise cease functioning inside the tumor.
Virally based cancer immunotherapy
Viruses whose replication and cytotoxicity is restricted to tumor cells, have proven helpful in treating some tumors ( [33] ). However, it is difficult to infect the majority of cells in a tumor. Fortunately an important similarity between a tumor partially infected with an immunotherapeutic virus, and many naturally occurring viral infections, suggests that direct cytolytic function is not necessary for a virus to fully eliminate a tumor. In both a natural viral infection, and in a tumor partially infected with a virus, only some cells display both disease-related Ags and danger signals; others are compromised. Although our immune systems are typically unable to sustain an attack against tumors after they fully evolve suppression of danger signals or Ag display, they can mount an effective attack against viral infections, even when only some infected cells display danger signals and viral peptides.
Such success against covert viral pathogens suggests that infection of tumors with an appropriately programmed and selective immunotherapeutic virus may succeed in complete elimination of tumor, if the virus upregulates danger signals in a sufficient fraction of living tumor cells. The virus’ selectivity toward tumor cells would have to extend not only to viral replication, but also to danger-signal upregulation; otherwise T cells that compare infected and non-infected cells could not sense a correlation between danger signals and TAAs.
The original danger model would have dictated the use of a lytic, as opposed to non-lytic, immunotherapeutic virus to treat tumors—so that TAAs, freed from tumor cells necrotically destroyed by the virus, would be captured and presented by APCs. Such presentation by dendritic cells is very useful to tumor immunotherapy. For better or worse, any non-lytic immunotherapeutic virus that upregulates danger signals in its host cells, will quickly cause its host cells to be killed by the immune system, probably in a variety of ways, at least some of which are likely to result in TAAs being presented by dendritic cells along with potent costimulation. Therefore, even a non-lytic virus chosen for immunotherapy targeted directly at T cells, will also benefit from the great and synergistic advantages of dendritic-cell immunotherapy.