Approximately 25% of all people who are born today will ultimately die of infection1, and most of these individuals will die before they reach reproductive age. Infection therefore constitutes a tremendous selective pressure in humans, as in most multicellular species on earth. It has impelled the evolution of diverse resistance mechanisms, and to a large extent has shaped our genomes. Here we consider the mechanisms of host resistance to viruses, which we regard as particularly challenging adversaries in several respects. Viruses can be as virulent as any microorganism, but probably evolve faster than protozoa, fungi or bacteria. They use large numbers of host-encoded proteins for their activities, and as obligate parasites, can seek no productive refuge in the extracellular environment — they must at all times cope with immunity in one species or another.

To understand antiviral defences, it is useful to view them from an evolutionary perspective. Some survival strategies evolved long ago, and are represented across highly divergent species (for the present discussion, across insects and mammals). Some proteins that are required for surviving viral infection are used to combat many other microorganisms as well. In mammals, for example, myeloid differentiation primary-response gene 88 (MyD88), a Toll-like receptor (TLR) adaptor molecule, is required for effective resistance to herpesviruses, Toxoplasma gondii and other organisms that have few obvious features in common. By contrast, some defences that evolved recently are matched against specific microorganisms and may operate only within a single host species. For example, the Ly49H receptor expressed by mouse natural killer (NK) cells seems to recognize only mouse cytomegalovirus (MCMV), and does so only in some strains of mice, as the receptor-encoding gene has been deleted in other strains. More tentative defences are also apparent. For example, mutational abrogation of the human CC-chemokine receptor 5 (CCR5) protein offers strong protection against infection with HIV, but the most common protective allele has not been driven to fixation in humans. And it is likely that many other potential resistance mechanisms remain to be exploited in mammals.

Some protective antiviral systems are cell autonomous, whereas others depend on multiple specialized cell types that interact both with the infected cells within the host and also with one another. In mammals, the response to viral infection overlaps substantially with the response to bacteria, using some of the same sensing, signalling and effector mechanisms. In insects, there is a greater reliance on cell-autonomous defence, and new advances must be made in defining the pathways that are involved.

Recently, forward and reverse genetic analyses (Fig. 1) have unveiled a new understanding of how the host perceives viral infection and acts to counter it. This Review discusses selected mechanisms by which the host counteracts viral infection, as well as some of the countermeasures that have swiftly evolved in viruses to preserve their advantage. Parallel studies in insects and in mammals have cast antiviral immunity in relief, telling us which defensive strategies are ancient and which are new.

Figure 1: Principles and applications of forward and reverse genetic analysis.
figure 1

a | Poorly understood phenomena form the basis for all biological inquiry, and in the present case, that phenomenon is the inherent resistance of multicellular organisms to viral infection. The primary tool of investigation in reverse genetics is the hypothesis, whereas the primary tool of investigation in forward genetics is the phenotype. Scientists using reverse genetics test hypotheses by observing the effect of targeted gene modification (usually knockout). Scientists using forward genetics seek to find the molecular change that causes phenotypes. The two approaches are highly complementary. Each can be extended within its own methodological sphere (that is, with a fresh round of mutagenesis to create new phenotypes, perhaps with modification of the background or the screen, or with a fresh set of gene-targeting experiments). Ultimately, both approaches yield new questions for investigation. b | Both forward and reverse genetic methods can yield susceptibility and resistance mutations. Forward genetic methods provide insight into the workings of the innate immune system as it presently exists; reverse genetic methods may tell us how immunity might be made even stronger, and can disclose essential needs of the pathogen that are not essential for the host. Either type of mutation may potentially point to 'host-oriented' therapeutics.

TLRs and viral recognition

Nucleic-acid sensing by TLR3, TLR7, TLR8 and TLR9. TLRs are comprised of an ectodomain of leucine-rich repeats (LRRs), a transmembrane domain and a cytoplasmic domain known as the TIR (Toll/interleukin-1 receptor (IL-1R)) domain. TLR4 can sense protein components of divergent viruses2,3,4. However, the nucleic-acid-sensing TLRs (TLR3, TLR7, TLR8 and TLR9) have a more general protective role against viruses.

TLR3, TLR7, TLR8 and TLR9 reside within the endoplasmic reticulum (ER) and cytoplasmic vesicles, such as endosomes, and recognize nucleotides that are derived from microorganisms. Double-stranded RNA (dsRNA) and its synthetic analogue, polyinosinic–polycytidylic acid (polyI:C), are recognized by TLR3, viral GU-rich single-stranded RNA (ssRNA) is recognized by TLR7 in mice and TLR8 in humans, and bacterial or viral unmethylated DNA with CpG motifs is recognized by TLR9 (Refs 58). Although it is well established that polyI:C stimulates TLR3 to activate the interferon-β (IFNβ) promoter, the role of TLR3 in the host defence against most viruses is still controversial. TLR3 is involved in the cross-presentation of antigens from virus-infected cells through the engagement of virus-derived RNAs9 — that is, RNA from infected cells is detected by a population of responding cells that need not be directly infected. In addition, polyI:C-induced IL-12p40 production in vivo depends on the presence of TLR3 (Ref. 5), as does host defence against MCMV10. Nevertheless, it was also shown that Tlr3−/− mice were more resistant to infection by several other viruses, such as West Nile virus and influenza virus, suggesting that TLR3-mediated inflammatory responses were harnessed by these viruses for establishing infection11.

TLR7 and TLR9 have a crucial role in the recognition of viruses by plasmacytoid dendritic cells (pDCs). pDCs seem to have evolved as specialized TLR-dependent virus-sensing cells that are endowed with the capacity to produce high amounts of type I IFNs in response to viral infection. This specialization may depend on high levels of constitutive IFN-regulatory factor 7 (IRF7) expression, and perhaps on the long retention of TLR9 ligands within the endocytic pathway, which might facilitate encounters between TLRs and their ligands. Most likely, other adaptations are also required for the evloution of pDCs that produce much (but not all) of the type I IFN that is detected in the plasma during the course of certain viral infections despite the relatively low frequency of these cells in the blood and peripheral lymphoid organs12,13. As discussed in detail later, different cell types use different receptors to detect viral infection, and pDCs provide only one example; it should not be thought that they are unique in their ability to manage viral infection.

TLR3, TLR7, TLR8 and TLR9 are believed to signal from acidified endosomal compartments because numerous drugs that block acidification also inhibit signalling by these TLRs14,15,16,17. Mutations in the multiple-transmembrane-spanning protein UNC93B can entirely block TLR3, TLR7, TLR8 and TLR9 signalling18,19. UNC93B, which is located mainly in the ER, physically interacts with these TLRs, as well as with TLR13, but not with other TLRs20. The presence of UNC93B may be required for these TLRs to gain access to the endosomes, or it may be an essential part of the receptor complex. Interestingly, UNC93B is also required for effective cross-presentation of antigens, and as such, is a protein with direct connections to both innate (TLR signalling) and adaptive (T-cell activation) immune functions.

Adaptors and downstream consequences of signalling. Following stimulation with polyI:C, TLR3 triggers a signalling cascade through the adaptor protein TRIF (TIR-domain-containing adaptor protein inducing IFNβ)21,22 (Fig. 2). TRIF associates with downstream signalling molecules, including tumour-necrosis factor (TNF)-receptor-associated factor 3 (TRAF3), TRAF6 and receptor-interacting protein 1 (RIP1). TRAF6 and RIP1 activate nuclear factor-κB (NF-κB)23,24. In addition, other molecules appear to be required to connect TRIF to the activation of NF-κB, because an ENU-induced mutation known as Feckless abolishes TLR3-dependent activation of NF-κB25. On the other hand, TRAF3 is responsible for inducing the production of type I IFNs26,27. TRAF3 recruits and activates two IKK (inhibitor of NF-κB (IκB) kinase)-related kinases, TBK1 (TANK-binding kinase 1) and IKKi (inducible IKK; also known as IKKɛ). Phosphorylation of IRF3 and/or IRF7 by these kinases induces the formation of homodimers. IRF3 and/or IRF7 then translocate into the nucleus and bind to the IFN-stimulated response elements (ISREs) in the genes that encode type I IFNs and in a set of IFN-inducible genes, resulting in their expression (reviewed in Ref. 28).

Figure 2: Signalling through TLR3, TLR7, TLR8 and TLR9 in response to endosomal nucleic acids of viral origin.
figure 2

The Toll-like receptors (TLRs) that sense nucleic acids can operate in non-infected cells of many types to detect the production of infection in other cells. Following the recognition of viral double-stranded RNA (dsRNA), single-stranded RNA (ssRNA) or CpG-containing DNA by TLR3 or TLR7, TLR8 and TLR9 that are expressed in the endosome, signalling proceeds through TIR (Toll/interleukin-1 receptor (IL-1R))-domain-containing adaptor protein inducing interferon-β (TRIF) or myeloid differentiation primary-response gene 88 (MyD88), respectively. UNC93B is a multiple-transmembrane-spanning protein that is predominantly located in the endoplasmic reticulum (ER), but is known to associate with these endosomal TLRs and to be required for them to signal. TRIF, through the recruitment of tumour-necrosis factor receptor (TNFR)-associated factor 6 (TRAF6) and receptor-interacting protein 1 (RIP1), as well as TANK-binding kinase 1 (TBK1) and inducible IκB (inhibitor of nuclear factor-κB (NF-κB)) kinase (IKKi) activate interferon (IFN)-regulatory factor 3 (IRF3) and NF-κB. MyD88 recruits TRAF6 and IL-1R-associated kinase (IRAK) and activates IRF7 and NF-κB. TLR4 (not shown) also detects viruses, signalling in response to specific virally encoded proteins through MyD88, TRIF and/or TRAM (TRIF-related adaptor molecule). NF-κB, IRF7 and IRF3 translocate to the nucleus to induce the transcription of genes encoding cytokines such as TNF, IL-6 and type I IFNs.

TLR7- and TLR9-dependent signalling requires the TIR-domain-containing adaptor MyD88 but not TRIF. TLR7 and TLR9 have a crucial role in the recognition of viruses by pDCs. Following exposure to ssRNA and CpG-containing DNA, MyD88 forms a complex with IL-1R-associated kinase 4 (IRAK4), IRAK1, TRAF3, TRAF6, IKKα and IRF7, and this complex is recruited to the TLR27,29,30,31. This signalling complex is essential for the induction of type I IFNs and pro-inflammatory cytokines by activating IRF7 and NF-κB, respectively. Among the proteins in the complex, IRAK1, TRAF3 and IKKα are responsible for the activation of IRF7 through phosphorylation, whereas MyD88 and IRAK4 regulate both IRF7 and NF-κB. pDCs that lack both TBK1 and IKKi can produce IFNα in response to TLR9 stimulation, indicating that pDCs use a unique signalling pathway for the production of IFNs32. Osteopontin, long known as the principal phosphorylated glycoprotein of bone and more recently recognized for its importance in T helper 1 (TH1)-cell immune responses, co-localizes with MyD88 in the cytoplasm, and is required for the production of IFNα in response to TLR9 stimulation33.

Cytoplasmic sensors and signalling pathways

Nucleic-acid sensing by RIG-I and MDA5. As TLRs are transmembrane proteins with pathogen-recognition domains that do not face the cytoplasm, TLRs cannot recognize viral components that are present in the cytoplasm. In the course of an infection with an RNA virus, dsRNA is generated as an intermediate product of replication in an infected cell. Bidirectional transcription from dsDNA viral genomes also leads to abundant levels of dsRNA in the cytoplasm. Furthermore, RNAs from some viruses are 5′-triphosphorylated and uncapped, whereas host mRNA is capped to prevent recognition by the innate immune system34,35,36. A cytoplasmic protein, retinoic-acid-inducible gene I (RIG-I), was shown to be a viral RNA detector, which induces the production of type I IFNs37. RIG-I is comprised of two caspase-recruitment domains (CARDs) at the N-terminal and a C-terminal DEXD/H-box RNA-helicase domain (Fig. 3).

Figure 3: Signalling through cytoplasmic helicases leads to the activation of both NF-κB and IRF3.
figure 3

Viral double-stranded RNA (dsRNA) in the cytoplasm is detected by one of two cytoplasmic helicases, melanoma differentiation-associated gene 5 (MDA5) or retinoic-acid-inducible gene I (RIG-I). RIG-I also detects 5′-triphosphate single-stranded RNA (ssRNA). Interferon-β (IFNβ)-promoter stimulator 1 (IPS1; also known as mitochondrial antiviral signalling protein (MAVs), virus-induced signalling adaptor (VISA) and caspase-recruitment domain (CARD) adaptor inducing IFNβ (CARDIF)) is a central target of both MDA5 and RIG-I. Through the recruitment of tumour-necrosis factor receptor (TNFR)-associated factor 3 (TRAF3), IPS1 activates FAS-associated death domain (FADD). FADD then associates with pro-caspase-8 or pro-caspase-10, resulting in its cleavage. The death-effector domains (DEDs) of caspase-8 or caspase-10 activate nuclear factor-κB (NF-κB), but not IFN-regulatory factor 3 (IRF3). However, the interaction between TRAF3 and TANK-binding kinase 1 (TBK1) or inducible IκB (inhibitor of NF-κB) kinase (IKKi) leads to IRF3 and IRF7 activation. Tripartite-motif-containing 25 (TRIM25) catalyses lysine-63-linked ubiquitylation of RIG-I, which strongly enhances RIG-I signalling. DD, death domain; JEV, Japanese encephalitis virus; VSV, vesicular stomatitis virus.

RIG-I forms a family with melanoma differentiation-associated gene 5 (MDA5) and LGP2 (both of which also have a C-terminal RNA-helicase domain) based on the high similarity of the helicase domain between these proteins, and this family is known as the RIG-I-like receptor (RLR) family38,39. RLRs interact with dsRNAs through the helicase domain, and the CARDs are responsible for triggering downstream signalling cascades. As LGP2 lacks a CARD, it is suggested that it to functions as a negative regulator of RIG-I and MDA5 signalling39,40. Recent analysis of RIG-I-deficient mice and MDA5-deficient mice revealed that RIG-I and MDA5 recognized different RNA viruses41,42,43. RIG-I-deficient cells did not produce type I IFNs in response to certain RNA viruses, including paramyxoviruses, influenza virus, Japanese encephalitis virus and vesicular stomatitis virus (VSV). By contrast, MDA5-deficient cells did not produce type I IFNs in response to picornaviruses, including encephalomyocarditis virus (EMCV) and Theiler's virus. Consistent with these abrogated IFN responses, RIG-I-deficient mice and MDA5-deficient mice were highly susceptible to infection with VSV and EMCV, respectively. These results indicate that RLRs have a crucial role in the host defence against infection by RNA viruses in vivo. Furthermore, RIG-I and MDA5 recognize different types of RNA — in vitro-transcribed dsRNAs and polyI:C, respectively.

In addition to dsRNAs, 5′-triphosphate ssRNAs are also detected by RIG-I and induce type I IFN production44,45. It was shown that cells that were infected with EMCV but not with influenza virus generated dsRNA in the cytoplasm. When 5′-triphosphate ssRNA or the whole influenza virus genome was treated with a phosphatase to remove the 5′-triphosphate group, the RNAs failed to induce type I IFNs, suggesting that the recognition of influenza virus by RIG-I is mediated through 5′-triphosphate genomic ssRNA. Although RIG-I was shown to possess intrinsic helicase activity to unwind short dsRNA, further studies are required to clarify whether this catalytic activity is needed for the detection of 5′-triphosphate ssRNA46. Recently, it was reported that tripartite-motif-containing 25 (TRIM25) catalysed the lysine 63 (K63)-linked ubiquitylation of RIG-I47. This ubiquitylation strongly enhanced RIG-I signalling, and cells that lack TRIM25 showed severe impairment in IFNβ production in response to RIG-I stimulation. Recently, mice deficient in the RLR LGP2 were reported to show increased IFN responses to polyI:C stimulation and VSV infection, whereas the responses to EMCV were modestly impaired in these mice48. However, the relationship between LGP2, RIG-I and MDA5 is still controversial, and further studies will be required to clarify the role of LGP2 in RNA-virus recognition.

IPS1 as the signalling molecule. The CARDs of RIG-I and MDA5 are responsible for initiating the signalling cascades. RIG-I and MDA5 interact with the adaptor protein IFNβ-promoter stimulator 1 (IPS1; also known as mitochondrial antiviral signalling protein (MAVS), virus-induced signalling adaptor (VISA) and CARD adaptor inducing IFNβ (CARDIF)) through the homophilic interaction of CARDs49,50,51,52. Overexpression of IPS1 induces the activation of the IFN promoters, as well as the activation of NF-κB. Ips1−/− mice were defective in producing type I IFNs and pro-inflammatory cytokines in response to all RNA viruses recognized by either RIG-I or MDA5, indicating that IPS1 is essential for both RIG-I and MDA5 signalling53,54. IPS1 is present in the outer mitochondrial membrane, suggesting that mitochondria might be important for IFN responses, in addition to their roles in metabolism and cell death51. Downstream of IPS1, TRAF3 has been shown to be important for the production of IFNs. The C-terminal TRAF domain of TRAF3 associates with a TRAF-binding domain that is found in IPS1. TBK1 and IKKi are activated by a TLR3-dependent pathway, indicating that the signalling pathways that are triggered by TLR stimulation and RIG-I converge at the level of TBK1 and IKKi. In addition, the FAS-associated death domain (FADD) protein has been reported to be required for type I IFN production in response to dsRNA, although another report showed that virus-induced type I IFN responses were not impaired in Fadd−/− cells39,55. FADD interacts with pro-caspase-8, pro-caspase-10 and IPS1, and a FADD-dependent pathway is responsible for the activation of NF-κB downstream of IPS1 (Ref. 56) (Fig. 3).

Viral evasion mechanisms

It has been shown that various viruses evade host immune systems by targeting virus-recognition mechanisms and the IFN signalling, or the RNA interference (RNAi) pathways in insects (see later). RIG-I was reported to bind influenza virus non-structural 1 (NS1) protein, which inhibited RIG-I-mediated 5′-triphosphate RNA recognition. IPS1 is cleaved by a hepatitis C virus protease NS3/4A50,57,58. However, TLR7- and TLR9-mediated type I IFN production by pDCs was shown to be inhibited by respiratory syncytial virus and measles virus. Vaccinia virus A46R suppressed TLR3-mediated type I IFN responses by binding to TRIF59. These reports indicate that the TLR system, as well as RLRs, is targeted by viral proteins. As with other forms of innate immune sensing, it is plausible that the host immune system has maintained multiple virus-recognition mechanisms to ensure efficacy against a wide range of viruses, some of which might evade innate immune detection through one pathway yet remain vulnerable to detection through a different pathway.

The relatively large viruses of the Herpesviridae family have developed mechanisms to evade not only cell-autonomous immunity, but also the molecular interactions that lead to responses by specialized antiviral components of the innate and adaptive immune systems. NK cells provide the essential effector functions that lead to the clearance or control of viral infection during the first days following infection, and their activation depends on the recognition of 'induced-self' (proteins that are activated in the host cell in response to infection), 'missing-self' (downregulation of MHC proteins on infected cells in response to viral infection) or 'non-self' (induced by virally expressed proteins or foreign MHC molecules) by specific NK-cell receptors. The expression of m144 by MCMV, which encodes an MHC class I homologue, impairs NK-cell recognition of virus-infected cells, presumably by replacing 'missing-self', and engaging NK-cell inhibitory receptors60. MCMV-encoded gp40 also prevents cell-surface expression of retinoic acid early transcript 1 (RAE1; an induced-self ligand), and therefore impairs NK-cell recognition of virus-infected cells by means of the RAE1 receptor NKG2D (natural killer group 2, member D)61. In a similar manner, m145 of MCMV encodes a protein that downregulates the expression of murine ULBP-like transcript 1 (MULT1; another NKG2D ligand)62 and m155 downregulates the expression of H60 (another NKG2D ligand)63.

Adaptive immune activation requires the presentation of foreign antigen by MHC molecules and a co-stimulatory signal. MCMV selectively downregulates host MHC-class-I-molecule expression and/or activity by means of three proteins, encoded by the m04, m06 and m152 genes. gp34, encoded by m04, forms a complex with MHC class I molecules in the ER, is transported to the cell surface and acts there to inhibit cytotoxic T lymphocyte (CTL) activation64. gp48, the product of the m06 gene, binds to MHC class I molecules and redirects them to the lysosome for degradation65. m152 encodes gp40, which is involved in the retention of MHC class I complexes in the ER and Golgi compartments and thereby inhibits the development of virus-specific CTL activity66.

gp34, gp48 and gp40 specifically interfere with antigen presentation and are collectively known as VIPRs (viral proteins interfering with antigen presentation) or immunoevasins67,68. But MCMV can also block co-stimulation, and a protein that is encoded by m147.5 downregulates CD86 expression on antigen-presenting cells (APCs), such as myeloid DCs69, hindering the CD28-mediated signal required for lymphocyte expansion. Other mechanisms of adaptive immune activation are also targeted. m36 encodes a protein with anti-apoptotic functions in macrophages and binds to pro-caspase-8 and inhibits the death-receptor-mediated induction of apoptosis70. This not only allows protracted viral proliferation within the host cell, but may also prevent the generation of CTL responses through the activation of B220 lymphoid DCs, which efficiently present antigens in the context of cell death, driving a powerful CTL response71.

Cell-autonomous defences are also targeted by CMVs. m142 and m143 are both required for the inhibition of the IFN-inducible dsRNA-dependent protein kinase (PKR)-mediated host antiviral response72. In human cytomegalovirus (HCMV) infection, the production of virally encoded cytokine receptor antagonists, such as vIL10, inhibits immune-cell proliferation, the synthesis of inflammatory cytokines, such as IL-1α, IL-6, TNF and granulocyte/macrophage colony-stimulating factor (GM-CSF), and MHC-class-II protein expression, leading to reduced CD4+ T-cell activity73.

MCMV infection: cooperation among cells

The need for cellular specialization. Diverse defence mechanisms that evolved in mammals as viral-evasion mechanisms nullified individual defences in turn, and the simultaneous use of different resistance strategies that are commonly seen in innate immunity, are probably essential for the survival of the host. As alluded to earlier, innate antiviral responses in mammals are not always cell autonomous, but instead may involve cooperation between two or more cell types with specialized antiviral potential. The collaboration between multiple classes of DCs and NK cells, occurring independently of cell-autonomous defence, is required for the successful containment of MCMV infection and provides an excellent example of such collaboration (Fig. 4).

Figure 4: Central roles of NK cells and DCs in the defence against MCMV.
figure 4

Murine cytomegalovirus (MCMV) infects several cell types including epithelial cells and conventional dendritic cells (DCs), but not plasmacytoid DCs (pDCs). Infected cells express the viral protein m157, which is recognized by the natural killer (NK)-cell-activating receptor Ly49H. Whether directly infected or not, conventional DCs sense nucleic acids that are produced by MCMV through Toll-like receptor 3 (TLR3) and TLR9, leading to the production of type I interferons (IFNs), interleukin-12 (IL-12) and IL-18. These cytokines induce signals through receptors that are expressed by NK cells, acting in the case of IL-12 and type I IFNs through signal transducer and activator of transcription (STAT) molecules. The viral DNA is also sensed by the pDC, the main producer of type I IFNs. The recognition of m157 and the production of type I IFNs are required to trigger full activation of NK cells. Activated NK cells produce IFNγ and expel lytic granules that contain granzymes and perforin into a synaptic cleft that is formed between the NK cell and its infected target cell, lysing the target cell. IFNγ contributes to the further activation of conventional DCs and other immune cells. MCMV infection is thus sensed by both DCs and by NK cells directly, and cooperation between the two cell types is required for early defence against infection. Mutations that affect sensing (for example, through TLR3 and TLR9), signalling (for example, through IFNs or IL-12 and IL-18), effector function or homeostasis can be lethal in the context of infection. MyD88, myeloid differentiation primary-response gene 88; TRIF, Toll/IL-1-receptor-domain-containing adaptor protein inducing IFNβ.

MCMV is known to infect vascular endothelial cells as its first target after inoculation, but later it can infect many epithelial cell types as well. These cells cannot contain the virus through cell-autonomous defences. The central role of NK cells in the response to MCMV was first established by antibody-depletion studies, which showed that C57BL/6 SCID mice (severe combined immunodeficient mice), which are normally resistant to MCMV, become highly vulnerable to infection when NK cells are removed by an NK1.1-specific antibody74 or by an asialo-GM1-specific antibody75,76. Subsequently, a quantitative trait locus known as Cmv1 was found to account for most of the phenotypic difference between BALB/c (susceptible) and C57BL/6 (resistant) mice77. This locus was positionally cloned and shown to encode the NK-cell activating receptor Ly49H78,79. Ly49H recognizes a protein encoded by the viral gene m157, a distant homologue of MHC class I proteins80,81. When activated by m157 (non-self), Ly49H signals by recruiting the immunoreceptor tyrosine-based activating motif (ITAM) adaptor protein DAP12 (also known as KARAP or TYROBP), which is used by proteins of the TREM (triggering receptor expressed on myeloid cells) superfamily82.

DAP12 signalling leads to the recruitment of the spleen tyrosine kinase (SYK) family, including SYK and ζ-chain-associated protein kinase of 70 kDa (ZAP70)83,84, which triggers an increase in intracellular calcium concentration, cytotoxicity and the induction of IFNγ expression. A second activation signal is supplied by type I IFN receptor (IFNAR) signalling, and mutations in IFNAR or in its associated signalling molecule STAT1 (signal transducer and activator of transcription 1) cause even more severe impairment of resistance to MCMV than a mutation in Ly49H — death occurs within 4 days rather than 5 days with identical doses of the virus85. One source of type I IFN is pDCs, and the activation of pDCs to produce IFN is at least largely TLR dependent, as mutations that affect TLR9, TLR3, MyD88, TRIF or UNC93B cause marked impairment of the type I IFN response to infection by MCMV in vivo10,18,86. However, it is not clear whether the pDCs themselves make a non-redundant contribution to defence against MCMV, as ablation of pDCs (for example, by a hypomorphic allele of Ikaros )87 leads to approximately 10-fold less IFNα production in vivo following MCMV challenge, and approximately fourfold higher viral burden 1.5 days post inoculation. Moreover, antibody-mediated depletion of pDCs does not make MCMV infection lethal in mice that are normally resistant, nor does it cause a significant increase in viral burden in the spleen after infection88. It is probable that other cytokines, notably IL-12 and IL-18, which are produced by CD8+ DCs, drive the expansion of Ly49H+ NK cells, which in turn produce IFNγ89. This may compensate for the relative deficit of type I IFN that is produced by pDCs.

The mechanism that couples type I IFN and DAP12 signalling to the activation of the effector function of NK cells is not yet clear. However, it is certain that NK cells must secrete the contents of their stored granules, which contain perforin, granzyme A and granzyme B, in close proximity to infected cells in order to kill them, and mutations that affect perforin or both granzyme A and granzyme B impair killing and viral clearance90. The molecular machinery for NK-cell granule exocytosis has been elucidated in part by random germline mutagenesis. Moreover, a general estimate of the number of genes that have non-redundant function in the resistance to MCMV has been offered based on the frequency with which susceptibility is induced by germline mutations85, and shows that the MCMV 'resistome' encompasses hundreds of genes. Many of these genes may also confer resistance to other microorganisms.

NK-cell exocytosis machinery depends on systems that are partially shared with CD8+ T cells, neutrophils, platelets, neurons, melanocytes and perhaps other cells. Cytolytic vesicles undergo polarization to the side of the cell from which their contents will be ejected, a process that requires their transport by microtubules, which in turn requires the assembly of the microtubule-organizing complex (MTOC) from one of the two centrioles present in all diploid cells (reviewed in Ref. 91). The lysosomal trafficking regulator (LYST) protein is needed for granules to form properly. The exact role of LYST is still unknown. Overexpression of LYST in fibroblasts induces the production of unusually small lysosomes, suggesting that LYST is involved in lysosome fission, as well as other events within the cell such as nuclear accumulation of phosphatidylinositol-4,5-bisphosphate (PtdIns(4,5)P2)92. In patients who have LYST mutations (which result in Chediak–Higashi syndrome), lytic granules polarize at the immunological synapse but do not undergo exocytosis, suggesting that LYST may be involved in docking and/or fusion of the lytic granules with the plasma membrane93 and may interact with SNARE proteins (soluble-N-ethylmaleimide-sensitive-factor accessory-protein receptor proteins) to permit fusion94,95.

RAB27a is required for polarized vesicles to leave the microtubules and become tethered to the plasma membrane91, and mutations in RAB27A cause Griscelli syndrome96, a disorder of pigmentation and immunity with features of haemophagocytic lymphohistiocytosis (HLH) — a fatal inflammatory disorder in which macrophages and CTLs proliferate dramatically. UNC13D (the human orthologue of which is MUNC13-4) also serves a priming function in granule exocytosis by making tethered granules competent to fuse with the membrane. In humans and mice, mutations of these orthologous genes result in HLH97,98. In mice, HLH does not occur spontaneously, but is clearly triggered by some (although not all) viral infections; in particular, the arenavirus lymphocytic choriomeningitis virus (LCMV)98. Syntaxin-11 (a SNARE protein) causes fusion to occur in response to changes in cytoplasmic calcium concentrations99,100, and mutations that destroy this protein also cause HLH.

Mutations in most (and probably all) of the exocytic components that are listed above cause viable phenotypes in which there is profound impairment in the resistance to infection by MCMV. At least two of these proteins, RAB27a and LYST, are also partially required for melanosome exocytosis, as their destruction causes relative hypopigmentation. Melanosome and neutrophil machinery are shared to a greater extent, so that many mutations that are associated with Hermansky–Pudlak syndrome cause both neutrophil and melanocyte dysfunction, although they do not affect NK-cell function. NK cells and CTLs use an exocytic mechanism that, so far, seems to be identical.

The need for homeostasis in the context of an innate immune response. Some of the mutations that impair MCMV resistance operate at a 'post-effector' level. These mutations impair homeostatic mechanisms that safeguard the host from the detrimental effects of the innate immune response. For example, the elimination of KIR6.1, a protein which, together with sulphonylurea receptor 2 (SUR2), forms an ATP-sensitive channel for the efflux of potassium ions from smooth-muscle cells in the coronary arteries, leads to hypersusceptibility to MCMV infection. This is because cytokines that are produced by haematopoietic cells cause severe vasoconstriction of the coronary arteries during infection when the channel is absent, leading to myocardial infarction101. Mice that lack KIR6.1 are also hypersensitive to lipopolysaccharide (LPS), polyI:C and CpG-containing DNA. The cytokines that mediate LPS-induced lethality are not known, but the phenotype is not suppressed by STAT1 deficiency or by TNF deficiency. This orthologous channel is also expressed in the heart in Drosophila melanogaster, and interestingly, knockdown causes the flies to become hypersusceptible to infection by flock house virus (FHV), suggesting that both insects and mammals rely on a common homeostatic mechanism for protection during infection, mediated by the channel.

Not all potential resistance mechanisms have been exploited in mice. Some strains of mice that lack Ly49H, such as the MA/My strain, show robust resistance to MCMV, whereas BALB/c mice do not102. In this particular case, resistance appears to be based on a functional interaction between Ly49P (an NK-cell-activating receptor) and H2-DK expressed by the infected target cell103, so that Ly49P compensates for the lack of Ly49H. Hence, other response mechanisms certainly exist. Moreover, random mutagenesis has revealed several mutations that make mice highly resistant to MCMV, just as mutations that affect the sensing, signalling or effector functions of the DC–NK-cell axis make them hypersusceptible (C.E. and B.B., unpublished observations). The genes that are altered by these mutations remain to be established. However, it appears that there are many proteins that are dispensable as far as the mammalian host is concerned, yet are essential for the virus to operate as an effective pathogen. MCMV is endemic in wild mice just as HCMV is endemic in human populations, and neither virus normally causes serious disease. It therefore appears that the DC–NK-cell axis is sufficient to contain them. Other resistance mechanisms might be of little benefit to the host, and may remain latent until such time as a mutated virus defeats the primary containment mechanism, whereupon they will confer a strong survival advantage.

Evolutionary perspective: viral defence in flies

Inducible antiviral defence in flies. One hallmark of the response of flies to infection by bacteria and fungi is the strong and rapid induction of hundreds of genes. This inducible response leads to the secretion in the haemolymph of numerous antimicrobial peptides that help to control the infection. The expression of the genes that encode these peptides is controlled by two signalling pathways, the Toll and IMD (immune deficiency) pathways, which regulate different members of the NF-κB family of transcription factors, namely DIF (dorsal-related immunity factor) and Relish, respectively104. Drosophila C virus (DCV) is the best-characterized D. melanogaster virus105,106. Infection of flies with this picorna-like virus does not trigger the strong humoral response that is characteristic of bacterial and fungal infections. In particular, the antimicrobial peptides regulated by either the Toll (for example, drosomycin) or the IMD (for example, drosocin) pathways are not detected in the haemolymph of DCV-infected flies107. The difference between the response to bacterial and fungal infections on one hand, and viral infection on the other hand, was confirmed using genome-wide microarrays. Approximately 150 genes are induced by a factor of at least two within 24 or 48 hours after DCV infection, and approximately two-thirds of these genes are not induced by challenge with bacteria or fungi108. These results indicate that flies can detect virus infection and trigger signalling that leads to the expression of a set of genes that is distinct from those that are activated by other infectious agents. This difference most likely reflects the fact that flies use different strategies to counteract intracellular pathogens, such as viruses, and extracellular pathogens, for which a strong humoral immune response is more appropriate. Dissection of the promoter of the gene vir1, which is strongly induced by DCV infection, but not by bacterial or fungal infections, established that the virus-response element of the promoter coincided with a STAT-binding site108.

The Janus kinase (JAK)–STAT pathway in flies is much simpler than in mammals, as there is a single JAK encoded by the gene hopscotch, and a single STAT factor (STAT92E). The pathway is activated by the cytokine receptor Domeless, which is an orthologue of the gp130 subunit of the IL-6 receptor in mammals. Three structurally related cytokines, Unpaired-1 (Upd1), Upd2 and Upd3, activate Domeless109. Genetic analysis using hopscotch-mutant flies confirmed that the induction of expression of vir1, as well as several other virus-induced genes, depended on the presence of a functional JAK. The receptor Domeless is also involved in the induction of vir1 expression, suggesting that one of the three Upd cytokines is induced by virus infection, and triggers an antiviral response in non-infected cells (Fig. 5). Some virus-induced genes, such as Vago, do not contain STAT-binding sites in their proximal promoter, and remain fully inducible in hopscotch-mutant flies, pointing to the existence of JAK–STAT-independent mechanisms of antiviral defence108.

Figure 5: Schematic overview of antiviral defences in Drosophila melanogaster.
figure 5

Double-stranded RNA (dsRNA) from replication intermediates is recognized by Dicer-2 in virus-infected cells (here a fat-body cell; the fat body is the fly equivalent of the mammalian liver and white adipose tissue), which process them into small interfering RNAs (siRNAs). The RNA-binding protein R2D2 helps to release one strand of the siRNA duplexes (the passenger strand) and incorporates the other strand (the guide strand) into the RNA-induced silencing complex (RISC). The siRNA then guides the RISC towards RNA molecules that contain complementary sequences, which are degraded by the slicing enzyme argonaute 2 (AGO2). In addition to this intrinsic system of host defence, which targets viral RNA molecules with high specificity, Drosophila C virus (DCV) infection leads to the induction of the Janus kinase–signal transduction and activator of transcription (JAK–STAT) pathway in uninfected cells. This induction is mediated by the gp130-like receptor Domeless, suggesting that viral infection triggers the synthesis of cytokines of the Unpaired family. Follwoing ligand binding, Domeless activates the only JAK kinase in D. melanogaster, which is encoded by the gene hopscotch. This leads to tyrosine phosphorylation and activation of the D. melanogaster transcription factor STAT92E. STAT92E dimers translocate to the nucleus and activate gene transcription.

An evolutionarily conserved antiviral role of the JAK–STAT pathway is supported by independent studies in other invertebrate models. The induction of STAT DNA-binding activity in the mosquito cell line C6/36 has been reported following infection by the flavivirus Japanese encephalitis virus110. In addition, the large DNA virus white spot syndrome virus (WSSV; family Nimaviridae) that infects shrimps also induces STAT-binding activity in infected animals, and subverts it to enhance the expression of its immediate-early genes111.

Of note, some insect DNA viruses, such as the polydnaviruses, express IκB-like proteins that inhibit DIF and Relish expression by cells from D. melanogaster112. The existence of such viral suppressors suggests that NF-κB pathways may also participate in the control of viral infections. Dif-mutant flies have been reported to succumb rapidly to infection with the birnavirus Drosophila X virus (DXV), but loss-of-function mutants for other genes of the Toll pathway (spätzle, Toll, tube and pelle) resisted infection similar to wild-type flies113, so there is no conclusive evidence at this stage that the Toll pathway has a role in antiviral defences in flies.

The lack of induction of vir1 and several other genes in hopscotch-mutant flies correlates with an increased viral load in DCV-infected flies and a higher susceptibility to infection. This correlation indicates that at least some of the genes that are induced by DCV act to control viral amplification in flies108. The lack of sequence similarities between the genes induced by DCV infection in flies and known antiviral effectors in mammals (for example, 2-5(A) synthetase, Mx proteins and PKR) suggests that D. melanogaster developed original strategies to control viral infections. At this stage, the only known antiviral effector mechanism in flies is RNAi, but the genes that encode components of the RNAi machinery are not regulated by DCV infection, indicating that this mechanism is part of an intrinsic defence system in D. melanogaster.

RNA interference as an intrinsic antiviral defence mechanism. A key component of the RNAi pathway is the RNase III enzyme Dicer. The D. melanogaster genome encodes two Dicer enzymes, Dicer-1 and Dicer-2, which produce two types of regulatory RNA, small interfering RNAs (siRNAs) and microRNAs (miRNAs). Dicer-1 processes the precursors of miRNAs, whereas Dicer-2 recognizes long dsRNAs and cleaves them into siRNAs114. Dicer-2 has a crucial role in the host defence against RNA virus infections, as Dicer2-mutant flies are highly susceptible to infection by the dicistroviruses DCV and CrPV (cricket paralysis virus), the nodavirus FHV and the alphavirus Sindbis virus115,116,117. Increased susceptibility of the mutant flies correlates with an increased quantity of viral RNA in infected flies. The Dicer-2 RNase does not act autonomously on viral RNAs — the small RNAs generated by Dicer enzymes are incorporated in the RNA-induced silencing complex (RISC) and serve to guide it to cytoplasmic RNAs that contain complementary sequences. The targeted RNAs are then cleaved by the endoprotease argonaute (AGO) that is contained in the RISC (AGO1 in the case of miRNAs and AGO2 in the case of siRNAs). Consistent with a crucial role of RNAi in the control of viral infection, Ago2-deficient flies also show an increased sensitivity to infection with DCV, CrPV and DXV. R2D2, a cofactor of Dicer-2 that mediates the loading of siRNAs into the RISC, is also required to control the level of viral RNA in infected flies116,117,118. Interestingly, sequence comparison between the genomes of different D. melanogaster species indicate that Dicer2, Ago2 and R2d2 are among the 3% of the fastest-evolving genes in D. melanogaster, an observation that suggests that the rapid evolution and adaptation of viruses to their hosts exerts a strong positive-selection pressure on the RNAi genes119. In support of this hypothesis, three viral suppressors of RNAi have been identified in the genomes of FHV, DCV and CrPV. Of particular interest, these three molecules do not exhibit any sequence relationship and represent powerful tools to probe in greater detail the molecular mechanisms that mediate viral RNA silencing116,117,120,121.

Importantly, this model, which is based on genetic data, has been confirmed by molecular analysis. Indeed, siRNAs of viral origin can be detected in infected flies115,116, and these siRNAs can confer resistance to a viral challenge. Transgenic flies that express RNA1 from FHV, which encodes the viral RNA-dependent RNA polymerase, produce replicating viral dsRNAs that are recognized and processed into siRNAs by Dicer-2 — these flies are better able to resist a challenge with FHV compared with control flies. As expected, this protection is virus specific and the transgenic flies are not protected against challenge with DCV115. Overall, these genetic and molecular data point to the importance of the RNAi machinery to detect viral RNAs and degrade them (Fig. 5).

Apart from AGO1 and AGO2, flies have three other argonaute proteins, known as Piwi, Aubergine and AGO3, which are closely related in sequence. These molecules are associated with a third class of small RNAs, the recently discovered Piwi-associated interfering RNAs (piRNAs). piRNAs silence repetitive DNA elements in flies, including transposons and the endogenous retrovirus Gypsy. The provirus Gypsy is similar in structure and function to vertebrate proviruses, and contains a single promoter that regulates the expression of the viral Gag, Pol and Env genes122. To jump to a new location in the genome, Gypsy sequences must first be transcribed into mRNA. The endogenous replication of Gypsy is repressed by the genetic locus flamenco. This locus consists of a large number of truncated or defective transposon elements that produce hundreds of distinct piRNAs. These piRNAs then associate with Piwi proteins and guide them to cleave target RNAs that are expressed from endogenous retroviruses123,124. Importantly, Piwi proteins and piRNAs also participate in transposon silencing in vertebrates125,126,127.

RNAi was first shown to be a primordial antiviral host defence mechanism in plants. In the plant genetic model Arabidopsis thaliana, two Dicer-like (DCL) enzymes, DCL4 and DCL2, mediate the production of siRNAs from dsRNAs and are involved in antiviral reactions128. Apart from the fact that A. thaliana devotes two DICER genes to the control of viral infections, a key difference between plants and flies is that RNAi is systemic in plants, spreading from tissue to tissue. This systemic response involves cell-to-cell signalling that is mediated by DCL4-generated siRNAs of 21 nucleotides in length, coupled to the production of dsRNAs in non-infected tissues by host RNA-dependent RNA polymerases129. Thus, siRNAs can be amplified and act over long distances to mediate a protective antiviral state in non-infected cells. By contrast, RNAi appears to be cell autonomous in flies130. This difference may explain the need for a cytokine-mediated signalling mechanism that alerts non-infected cells of the infection in flies. The recent discovery that dsRNA is taken up by receptor-mediated endocytosis in D. melanogaster cells nevertheless raises the alternative possibility that viral-replication intermediates that are released from lysed cells are taken up by neighbouring cells, thereby inducing RNAi and preventing the spread of the viral infection131.

Comparison with mammalian antiviral defences. Overall, the current data from D. melanogaster point to the existence of two types of antiviral defence: an inducible response involving the induction of several hundreds of genes, and an intrinsic defence system based on RNAi. As in mammals, the inducible response relies in part on the JAK–STAT signal-transduction pathway. It is of particular interest to note that there is now evidence for three types of inducible response to infection by different microorganisms in flies, and that the signalling pathways that mediate these responses (the Toll, IMD and JAK–STAT pathways) bear striking similarities to the three main inflammatory pathways in mammals (the TLR–IL-1R, TNF and IFN–IL-6 pathways, respectively). An interesting aspect of the inducible response to virus infection is that some of the genes that are induced probably contribute to the pathology and the fatal outcome of the infection108,115. The recent discovery that the expression of the fly orthologue of SUR2 (see above) in the heart is required for resistance to FHV infection opens the way for a better understanding of the physiopathology of virus infection in flies101.

Despite the evolutionary conservation of the signalling pathways, the effector mechanisms appear to be different in flies and mammals, even though RNases participate in the control of the levels of viral RNAs in infected cells in flies (Dicer-2) and mammals (RNaseL). One intriguing question is the apparent absence of RNAi from the arsenal of antiviral defences in mammals. Indeed, mammals only have one Dicer gene (DICER), which is essential for the production of miRNAs and is required for development, similar to Dicer-1 in flies. Interestingly, miRNAs in mammals regulate signalling in immune cells and can affect viral RNA levels in infected cells. For example, miR32 in human cells is a negative regulator of the retrovirus primate foamy virus type 1 (PFV1)132, whereas miR24 and miR93 target the VSV genome. The relevance of this finding is illustrated by the increased susceptibility to VSV infection of a strain of DICER-deficient mice133. In spite of this requirement for DICER in the control of some viral infections in mammals, viral RNAs in mammalian cells do not appear to be recognized and processed into siRNAs, pointing to an important difference between flies and plants134,135. The absence of RNAi as an antiviral mechanism in mammals may reflect the fact that on one hand it can easily be circumvented by viruses (as shown by the potent RNAi suppressor B2 encoded by FHV, one of the simplest viruses known), and on the other hand its exquisite specificity that is based on base pairing of complementary nucleotide sequences has been achieved by other means by the adaptive immune system in mammals.

Summary and inferences

In the broadest sense, we can say that viral infections have led to the development of remarkably effective defences in both mammals and insects, some of which are clearly ancestral and some of which are not. We have much to learn from all antiviral mechanisms, both for pragmatic reasons (given the threat that viruses represent) and because viruses have clearly shaped our genomes in an impressive manner. As discussed here, a large percentage of the mammalian genome seems dedicated to the defence against viral infection. Forty-five percent of the mammalian genome is ultimately of viral origin136 although much of this DNA has propagated through replication in the germline without new infection per se. Nonetheless, it is obvious that viruses have shaped our genome in more ways than one.

Resistance to viral infection comes at a definite cost. In mammals, the cellular systems that confer protection against viruses are also capable of causing autoimmunity. This may be seen as a reflection of three facts. First, viruses have imposed a need to distinguish between host nucleic acid and foreign nucleic acid — a challenge that has been met by mammals in the large part, but not completely. Second, viruses certainly contributed to the evolution of adaptive immunity, upon which autoimmunity is predicated. And third, some of the elements of innate immunity that support autoimmune disease (among them the |IFNs and cells that produce them) evolved largely to combat viruses.

In evolution, the most tentative antiviral defences, which may be highly effective against a specific agent in a given species, may also be common. It is possible that much of our resistance to viral infection depends on such 'works in progress'. It appears that strong resistance to specific viral infections can be created in the laboratory with ease through the use of a mutagen, and it is even possible that some such mutations will have protective efficacy against many different viral infections. Although most antiviral drugs that are currently in use operate directly against virally encoded proteins, there is a sound genetic basis for the implementation of a 'host-oriented' pharmacological approach to viral therapy, as it is clear that specific genomic changes are well tolerated by mammals, yet create strong resistance to viral proliferation.