INTRODUCTION
HIV-1 is a chronic infection that persists despite the concerted action of the innate and acquired immune responses. A key component of innate immunity is the interferon (IFN) response and HIV-1 replication is substantially inhibited by type I IFNs both
in vitro (
1) and
in vivo (
2). A number of IFN-stimulated genes (ISGs) such as APOBEC3G, TRIM5α, and tetherin have been reported to attenuate retroviral replication
in vivo (
3–5), and SAMHD1 is likely to be similarly important in this regard (
6,
7). However, these factors are all either evaded in their natural hosts or antagonized by viral accessory genes (
8,
9). Numerous other ISGs, such as CNP, ZAP, and MOV10, have been reported to inhibit HIV-1 replication
in vitro (
10–12). However, none of these factors are likely responsible for the strong IFN-mediated inhibition of HIV-1 infection observed in most human primary cells and some cell lines, particularly during the early steps of the replication cycle (
6,
7,
10–13).
Myxovirus resistance (Mx) proteins are a family of dynamin-like GTPases first identified for their ability to confer resistance to lethal doses of influenza A virus (
14,
15). Most mammals encode two paralogous Mx proteins, Mx1 and Mx2 (sometimes referred to as MxA and MxB). The Mx1 proteins have been reported to exhibit activity against viruses from a variety of families, while the Mx2 lineage was previously thought to be limited to cellular functions (
16), despite being strongly induced by IFN. Recently, we (and others) found that Mx2 is capable of efficiently inhibiting the early steps of HIV-1 infection
in vitro (
17–20). Mx2 could therefore contribute to the IFN-mediated suppression of HIV-1 replication that is observed
in vivo (
2). Mx2 impedes the early steps of HIV-1 infection prior to chromosomal integration of proviral DNA (
18–20), perhaps by inhibiting nuclear import of HIV-1 DNA following reverse transcription (
18,
19). The mechanistic details of how inhibition occurs are currently unclear. However, the capsid (CA) region of
gag is a major determinant of Mx2 sensitivity, and a few single-amino-acid substitutions in CA have been reported to confer partial or complete escape from Mx2 activity (
18–20). Notably, host cyclophilins could be involved in Mx2 antiviral activity since several substitutions in the cyclophilin binding loop, including at the cyclophilin binding site, enable escape from Mx2 (
18–20). An interaction with host cyclophilins has been proposed as a requirement for Mx2 inhibition due to the ability of cyclosporine to rescue infection in the presence of Mx2 (
20).
Examining the spectrum of retroviruses inhibited by species variants of restriction factors has previously been of great value in uncovering the molecular details of antiretroviral activity. Because orthologous variants often inhibit different spectra of retroviruses, these analyses can define the determinants of viral sensitivity to, and host specificity of, antiviral activity. Here, we show that several, but not all, mammalian variants of Mx2 exhibit antilentiviral activity. Upon examining a panel of retroviruses, we noted that Mx2 antiviral specificity is largely confined to primate lentiviruses and that Mason-Pfizer monkey virus (MPMV), a betaretrovirus, is also partially sensitive to Mx2. In addition, using in vitro evolution and unbiased mutagenesis, we found that multiple localized substitutions in the C-terminal domain of CA can confer resistance to Mx2. Intriguingly, primate Mx2 variants exhibit species-dependent activity against O-group HIV-1 and some M-group HIV-1 CA mutants, indicating that divergent Mx2s have divergent antiretroviral specificities. Using chimeric Mx2 proteins and evolution-guided approaches, we demonstrate that a single residue close to the N terminus of Mx2 that has evolved under positive selection can determine this antiviral specificity.
DISCUSSION
Extant species are descended from ancestors that were each challenged by unique spectra of pathogens. Consequently, orthologous genome encoded antiviral defenses have been selected to inhibit divergent spectra of viruses. Here, we show that Mx2 exhibits species-dependent variation in antiviral specificity that is likely the result of diversity in ancient selective forces. All of the primate Mx2 proteins tested were capable of inhibiting HIV-1, whereas canine and ovine orthologs had no such activity. Notably, transfer of a small region of human Mx2 to canine Mx2 conferred anti-HIV-1 activity, indicating that canine Mx2 requires little modification to gain this function. It is possible that antiretroviral activity of Mx2 proteins was gained specifically in the primate lineage or that antiretroviral activity arose earlier and ovine and canine Mx2 proteins inhibit retroviruses that have yet to be tested. Further work is required to support or exclude either possibility. Interestingly, the ovine Mx2 protein did not localize to nuclear pores, and canine Mx2 proteins did so less robustly than did the primate Mx2 proteins. Thus, it is possible that robust nuclear pore localization emerged in Mx2 proteins alongside the acquisition of activity against primate lentiviruses.
Species variants of Mx2s have divergent antiviral specificities. Notably O-group HIV-1, HIV-1 P207S, and to a lesser extent HIV-1 E187V were inhibited by primate Mx2 proteins in a species-dependent manner. Although HsMx2 inhibited all of these viruses, AGMMx2 did not (but was able to inhibit an M-group virus, namely, HIV-1 NL4-3). In at least one instance, a single residue (residue 37) close to the N terminus of Mx2 was responsible for this specificity. Moreover, residue 37 was one of two residues that also governed specificity for group O HIV-1. Sequence analysis revealed that, similar to other genome encoded antiviral factors, the majority of Mx2 residues are evolving under purifying selection (
62,
63), which is likely driven by the requirement to maintain the structural and functional integrity of Mx2. Within this background, however, signatures of positive selection are apparent, particularly close to the N terminus of Mx2. Strikingly, residue 37, which we demonstrate to be one determinant of antiviral specificity, was identified as evolving under positive selection. This raises the possibility that the signatures of positive selection apparent near the N terminus of Mx2 could have been driven by antagonistic host-pathogen interactions. One such possible selective pressure could be the antilentiviral specifying activity of this region. We note, however, that it is unlikely that residue 37 is the only residue governing Mx2 antiviral specificity, since humanizing the first 29 residues of canine Mx2 also conferred anti-HIV-1 activity. In contrast, the reciprocal chimera, bearing the first 29 residues from canine Mx2 but retaining the original human residue at position 37, was rendered inactive against HIV-1. Thus, other determinants of antiretroviral activity or specificity exist within the first 29 residues of human Mx2. Further work is required to determine whether the first 29 residues solely specify a subcellular localization compatible with anti-HIV-1 activity or whether additional specificity/activity determinants reside within the extreme N-terminal region of Mx2.
Interestingly, the N-terminal region of Mx2, which determines antiviral specificity, is spatially distinct from the disordered loop 4 (L4) that has been uncovered as a major antiviral specificity determinant of the Mx1 protein (
62). Mx1 has been proposed to directly bind viral components through the stalk domain (
64,
65) and perhaps assemble into oligomeric arrays that stimulate GTPase activity and activate the antiviral function of the C-terminal effector domain (
66). A predicted structure of Mx2 places the N-terminal specificity determinant proximal to the “hinge-like” bundle signaling element (
67), spatially separated from L4. The distinct locations of these specificity determinants, considered alongside their differential dependence on GTPase activity (
18,
19), reinforce the notion that the antiviral activity of the Mx2 protein is mechanistically distinct from that of Mx1.
In addition to inhibiting the infection of primate lentiviruses, Mx2 also inhibited the betaretrovirus MPMV, albeit relatively weakly. Although it is unlikely that such a modest inhibition would be relevant in vivo, this observation highlights the possibility that Mx2 could inhibit divergent retroviruses. It may be that other untested mammalian Mx2 proteins possess potent activity against betaretroviruses or other retroviruses. Such divergent activities could be important in understanding the evolutionary pressures that have shaped the activity of primate Mx2 proteins.
TRIM5α is a well-characterized antiviral factor that, like Mx2, blocks retroviral infection at an early stage postinfection. The surfaces of CA that are bound by
Macaca mulatta TRIM5α have been extensively characterized for both MLV and HIV-1 (
28,
40,
41,
56,
68,
69). When the CA surfaces defining TRIM5α and Mx2 sensitivity are compared, similarities are apparent. Substitutions in both the cyclophilin binding loop and the N-terminal β-hairpin reduced sensitivity to both antiviral factors (
Fig. 5) (
28,
56). Indeed, screens of the same HIV-1 CA library identified the same M10I and P90T substitutions, which confer escape from Mx2, as changes that reduce sensitivity to rhesus TRIM5α (
28). Although multiple CA surfaces are important determinants of Mx2 sensitivity, not every change in these regions confers escape. For example, M10I and H87R substitutions desensitize HIV-1 to Mx2, whereas other changes at the same positions (M10L and the common polymorphism H87Q) do not. Clearly, both the position and the nature of the substitution in HIV-1 CA influence Mx2-sensitivity. Thus, the substitutions identified herein likely represent an underestimate of the total number and extent of changes that could confer reduced sensitivity to Mx2. Moreover, a comparison of M-group (NL4-3) and O-group (CMO2.5) CA sequences suggests that none of the substitutions described here account for the differential susceptibility of these viruses to AGMMx2. Again, this suggests that more, perhaps many more, positions in CA that are yet to be described govern sensitivity to Mx2.
Despite the similarities, there are also substantial differences between the sensitivity determinants for Mx2 and TRIM5α. In particular, two additional CA surfaces appear to be important for Mx2 sensitivity. First, one important CA surface thought to mediate interactions with the host is the cleavage- and polyadenylation-specific factor 6 (CPSF6) binding interface (
70,
71). Two substitutions in this region have been reported to facilitate escape from Mx2 (N74D [
19] and N57S [
18]). Although clearly important for Mx2 sensitivity, this binding interface is not generally considered important for interaction with TRIM5α. Second, although multiple CA residues influence sensitivity to TRIM5α, they overwhelmingly occur on the surface of the N-terminal domain of CA and, crucially, no major sensitivity determinants have been described in the C-terminal domain. Strikingly, both
in vitro evolution and the CA mutant library approaches described here identified multiple Mx2 sensitivity determinants in the C-terminal domain. In principle, the CA surface these substitutions define is exposed to the host cytosol in the intact conical core, although this surface would likely be more accessible to interactions with host factors following disassembly/uncoating of the conical CA lattice. We also note that sensitivity to Mx2 could be altered by CA mutations through one or more of several mechanisms. Mutants could change the CA stability, could alter the way CA is recognized by host restriction factors (including Mx2 itself), or could alter the route by which HIV-1 accesses the nucleus. Each of these processes could change the sensitivity to Mx2. It seems unlikely that a single host factor would contact the HIV-1 CA in so many spatially distinct locations, and so it is possible that some of the Mx2 escape mutants described here confer resistance by altering the stability, structure, or nuclear import pathway. In this regard, it is interesting that Mx2 antiviral activity is sometimes, but not always, dependent on host cyclophilins. CsA is known to produce cell-type-dependent effects with respect to multiple phenotypes. Indeed, sensitivity to TRIM5α exhibits a variable dependence on host cyclophilins. Variable expression levels of CypA have been suggested as a possible explanation for the variable effects of CsA on HIV-1 infection and sensitivity to restriction factors (
72).
As yet, we have been unable to identify any substitutions that confer efficient escape from Mx2
in vitro that also commonly occur in patient sequences. Moreover, all of the group M viruses tested thus far appear sensitive to inhibition by Mx2 (
18–20). This includes “transmitted founder” viruses (
18,
19) that have been proposed to be more resistant to inhibition by IFNs (
73,
74). Thus, Mx2 might not exert a sufficiently strong pressure
in vivo to drive the generation of CA escape mutants or maintain their continued presence. In light of this, sequences derived from cohorts of HIV-1-infected patients during acute infection or undergoing IFN-treatment might be required to observe frequent Mx2 escape mutants
in vivo. Conversely, the extreme constraints placed upon CA to maintain replicative fitness (
27) and evade both pattern recognition (
75,
76) and immune responses (
77–80) could simply result in Mx2-resistant viruses failing to thrive
in vivo. Consistent with this idea, most substitutions conferring escape from Mx2 have substantially reduced fitness.
Further work is required to uncover the mechanistic basis of how Mx2 inhibits HIV-1 infection in vitro and to ascertain whether this inhibition occurs in vivo. Clearly, any inhibition that does occur in infected individuals is unable to tip the balance in favor of the host and clear HIV-1 infection. Thus, a complete understanding of this inhibitory interaction, which is perhaps underexploited in natural settings, could pave the way to harnessing Mx2 antiviral activity in a therapeutic context.