Host and Viral Determinants of Mx2 Antiretroviral Activity

The species variants of Mx2 were cloned from the following cell lines: FrHK4 (Macaca mulatta NM_001079696.1), Vero (Chlorocebus aethiops [Vervet], KF925356), CPT-Tert (Ovis aries, KF925355), and MDCK (Canis lupus familiaris, XM_005638734.1). The cloning of human Mx2 was described previously (18). The different cell lines were stimulated overnight with 1,000 U of peginterferon alfa-2b/ml and total RNA was extracted using TRIzol (Life Technologies) in accordance with the manufacturer's instructions. This RNA was used as a template to generate cDNA (Superscript III; Life Technologies) in accordance with the manufacturer's instructions. Species variants of Mx2 were PCR amplified using a high-fidelity polymerase prior to insertion in the N-terminal myc-tagged doxycycline-inducible lentiviral vectors (LKO derivatives) using SfiI. Because no AGM Mx2 sequences were available in GenBank, we initially amplified AGMMx2 from cDNA using primers complementary to the human Mx2 5′ and 3′ untranslated regions. PCR-amplified DNA was cloned, and 12 clones were sequenced. A single clone was subsequently used as a PCR template to amplify AGMMx2 for insertion into the lentiviral expression system. Species variants of Mx2 were amplified using the following primers: MacMx2, 5′-CTC TCT GGC CGA GAG GGC CAT GTC TAA GGC CCA CAA GTC TTG GC-3′ and 5′-TCT CTC GGC CAG AGA GGC CTC AAT GGA TCT CTT TGC TGG AGA ATT G-3′; AGMMx2, 5′-CTC TCT GGC CGA GAG GGC CAT GTC TAA GGC CCA CAA GTC TTG GC-3′ and 5′-TCT CTC GGC CAG AGA GGC CTT CAA TGG ATC TCT TCG CTG GAG AAT TG-3′; OvMx2, 5′-CTC TCT GGC CGA GAG GGC CAT GTC GAT GTC CTA TAG GGC TTT GAA G-3′ and 5′-TCT CTC GGC CAG AGA GGC CTC ATA TGC TTT TGA AAT GGG GGA ATT C-3′; and CanMx2, 5′-CTC TCT GGC CGA GAG GGC CAT GTC TAA GGC CCA CGG TTC TC-3′ and 5′-TCT CTC GGC CAG AGA GGC CTT AAC TGA AGA ACA TGT AGA GTG CGC-3′.

Chimeric Mx2s were generated using overlapping PCR with the primers listed above and also the following primers: Can₅₈Hs, 5′-AGG TGG GAG TGC AGG ATC CGG TTT ATC TCG CCA AGG ACT TCA ACT TTC TCA CTT TGA AC-3′ and 5′-CTG ATT GTT CAA AGT GAG AAA GTT GAA GTC CTT GGC GAG ATA AAC CGG ATC CTG CAC TC-3′; Hs₅₈Can, 5′-GCA GAG AAG GAC GCT GCT TTC CTC GCC AAG GAG TTC AAT TTG CTG ACA CTG AAC C-3′ and 5′-CTG GGG GTT CAG TGT CAG CAA ATT GAA CTC CTT GGC GAG GAA AGC AGC GTC CTT CTC TG-3′; Can₂₉Hs, 5′-AGA AAA AGA AAT GAA TTT TGT CCA GCA GCA GCC GCC GCC ATT CGG CAC AGT GCC ACC AC-3′ and 5′-GAA ACA TCA TTT GTG GTG GCA CTG TGC CGA ATG GCG GCG GCT GCT GCT GGA CAA AAT TC-3′; Hs₂₉Can, 5′-GAA AAA AGA AAT GAA TTC CTT CCA GCA ACA GCC ACC GCC ATC TGA CGC GGC AGC AAG GC-3′ and 5′-CTT GCT GCC GCG TCA GAT GGC GGT GGC TGT TGC TGG AAG GAA TTC ATT TCT TTT TTC AG-3′; Mac₉₀AGM, 5′-CAG GAA ACA CGA GCC AGC CAA GGG CAA AGG GGC CCG AGA ACA ACC TGC ACA GCC AGT AC-3′ and the SfiI-ApaI fragment of MacMx2; AGM₉₀Mac, 5′-AGG AAA CAC GAG CCA ACC AAG GGC AAA GGG GCC CGA GAA CAA CCT GCA CAA CCA GTA CG-3′ and the SfiI-ApaI of AGMMx2; Mac₃₃₃AGM, 5′-GGG GCA CCG AGA AAA GCA TCA TTA ATG TGG TAC GGA ACC TCA CAT ACC CCC TCA AGA AG-3′ and 5′-TGT AGC CCT TCT TGA GGG GGT ATG TGA GGT TCC GTA CCA CAT TAA TGA TGC TTT TCT CG-3′; AGM₃₃₃MAC, 5′-GAA CCT CAC ATA CCC CCT CAA GAA GGG CTA CAT GAT CGT GAA GTG CCG GGG CCA GCA GG-3′ and 5′-GAT GAT CTC CTG CTG GCC CCG GCA CTT CAC GAT CAT GTA GCC CTT CTT GAG GGG GTA TG-3′; HsRTA, 5′-CCT TCC AGC AAC AGC CAC CGC CAT TCC GCA CAG CGC CAC CAC AAA TGA TGT TTC CTC C-3′ and 5′-GGA GGA AAC ATC ATT TGT GGT GGC GCT GTG CGG AAT GGC GGT GGC TGT TGC TGG AAG G-3′; AGMGTV, 5′-TCC AGC AAC AGC CAC CGC CAT TCG GCA CAG TGC CAC CAC AAA TGA TGT TTC CCC CAA AC-3′ and 5′-TTG GGG GAA ACA TCA TTT GTG GTG GCA CTG TGC CGA ATG GCG GTG GCT GTT GCT GGA AG-3′; MacG37R, 5′-CTT CCA GCA ACA GCC ACC GCC ATT CCG CAC AGT GCC ACC ACA AAT GAC GTT TCC C-3′ and 5′-GGG AAA CGT CAT TTG TGG TGG CAC TGT GCG GAA TGG CGG TGG CTG TTG CTG GAA G-3′; and AGMR37G, 5′-CTT CCA GCA ACA GCC ACC GCC ATT CGG CAC AGC GCC ACC ACA AAT GAT GTT TCC C-3′ and 5′-GGG AAA CAT CAT TTG TGG TGG CGC TGT GCC GAA TGG CGG TGG CTG TTG CTG GAA G-3′.

The tetracycline-inducible lentiviral expression vector (pLKO.dCMV.TetO/R) used in the present study was constructed by the transfer of a unique 2,150-bp SpeI/BamHI fragment (containing sequences related to the hPGK promoter, the Tet repressor, and internal ribosome entry site [IRES]) from pLKO.Tet-ON (21) into pLKO.dCMV.TetO (22) (replacing complementary promoter sequences upstream of the puromycin resistance gene). The unique SfiI site in this vector was destroyed using the following oligonucleotides 5′-CTC GAG ACT AGT GCC T-3′ and 5′-CAC TAG TCT CGA GAG G-3′. Myc-tagged HsMx2 was then PCR amplified (from pLHCX-myc-HsMx2) using the following primers 5′-CTC TCT GCT AGC ACC ATG GAG CAG AAA CTC ATC TCT GAA G-3′ and 5′-CTC TCT GTC GAC GGC CAG AGA GGC CTC AGT GGA TCT-3′ before being subcloned into the LKO-derived vector above using NheI and SalI sites (leaving the following sequence, located between the forward primer and the start codon of Mx2: AGG ATC TGG CCG AGA GGG CC). This generated the vector pLKOΔ-MycHsMx2-IP and subsequent open reading frames (ORFs) of interest were inserted directionally into this vector, using SfiI (retaining the N-terminal tag). These ORFs are situated downstream of a truncated human cytomegalovirus immediate early promoter containing a tandem TetO operator sequence (22), allowing for the inducible expression of proteins in the presence of doxycycline. The pLKOΔ-HA-Mx2-IP vectors were generated by digesting pLKOΔ-MycTagRFP-IP with NheI and EcoRV, and the HA-SfiI-HpaI-SalI-SfiI multiple cloning site was generated by using the oligonucleotides 5′-CTA GCC ACC ATG TAC CCA TAC GAT GTT CCA GAT TAC GCT AAG GCC GAG AGG GCC GTT AAC GTC GAC GGC CTC TCT GGC CGT CGA-3′ and 5′-TCG ACG GCC AGA GAG GCC GTC GAC GTT AAC GGC CCT CTC GGC CTT AGC GTA ATC TGG AAC ATC GTA TGG GTA CAT GGT GG-3′.

Human T cell lines were maintained in RPMI. Human HEK293T, macaque FrHK4, canine MDCK, and AGM Vero cells were all maintained in Dulbecco modified Eagle medium (DMEM). Ovine CPT-Tert cells are choroid plexus cells immortalized with the simian virus 40 T antigen and human telomerase reverse transcriptase and were kindly provided by David Griffiths (23). CPT-Tert cells were grown in Iscove modified Dulbecco's medium (IMDM). IMDM, DMEM, and RPMI medium was supplemented with 9% fetal calf serum. Variants of Mx2, Macaca mulatta TRIM5 (Mamu1 EF113914), Macaca mulatta TRIMCyp (EU157763), and TagRFP doxycycline-inducible cell lines were generated through transduction with modified LKO-derived lentiviral vectors, followed by selection in either 2 μg of puromycin (Sigma)/ml or 1 mg of G418 (Invitrogen)/ml. G418-resistant vectors were used in MT-4 TMZ R5 cells, and puromycin-resistant vectors were used in all other experiments. Expression was induced in these cell lines through an overnight treatment with 125 ng of doxycycline hyclate (Sigma)/ml prior to challenge with retroviruses or retroviral vectors. MT-4 TMZ R5 cells are derived from MT-4 cells. They stably express CCR5 and encode an LTR-GFP reporter construct (T. M. Zang and P. D. Bieniasz, unpublished data).

For replication-competent proviral clones we used pNHG (JQ585717), pNHGcapNM (JQ686832), pNL4-3 (M19921), and pCMO2.5 (24). All of the retroviruses and retroviral vectors used in the present study were generated through transfection of HEK293Ts using polyethyleneimine and titrated as described previously (25, 26). Briefly, viral stocks were titrated onto cells, and infected cells were quantified using flow cytometry. Titers were derived from at least three measurements at multiplicities of infection between 0.01 and 0.4. Cyclosporine (Sandoz) and Debio-025 (Novartis) were diluted in dimethyl sulfoxide and added to cultures at the time of infection, where indicated, at 5 μM.

For the CA mutant screen, purified plasmid DNA containing the green fluorescent protein (GFP)-encoding provirus NHGcapNM (accession JQ686832) and I2L, N5D, I6T, M10I, M10L, Q13H, A14S, I15V, N21S, S33C, Q50H, N57S, M68L, L69I, E75D, V83M, H87Q, H87R, P90T, I91T, I91V, Q95L, M96I, M96T, E98D, R100S, N121I, R132G, S146C, T148I, S149C, S149G, I150V, L151I, I153T, Y164F, R167Q, A177S, M185I, E187V, L190M, T200S, A204G, G208E, A209V, L211I, and T216A mutants thereof, were used to transfect HEK293T cells as described previously (27). Importantly, although some of these mutants have been classified as nonviable in spreading/replication assays (using <2% of the wild-type threshold [27]), all of the mutants were selected since they produce measurable titers of infectious virus in single-cycle transient-transfection/infection assays (the method used for the Mx2 screen). Filtered supernatants were titrated on doxycycline-treated MT-4 cells encoding HsMx2, MacMx2, AGMMx2, OvMx2, or CanMx2 in addition to untreated cells. Infected cultures were treated with dextran sulfate to limit infection to a single cycle (16 h postinfection) as described previously (27). All cells were fixed at 48 h postinfection and subjected to flow cytometry, and an infectious titer was calculated by counting GFP⁺ cells.

For the HIV-1 adaptation and replication assays, doxycycline-treated or untreated MT-4 cells (modified to express doxycycline-inducible HsMx2) were infected with HIV-1 (pNHG, JQ585717), and infection was monitored every 24 h by flow cytometry as described previously (10, 28). Filtered (0.45-μm pore size) culture supernatants were used to inoculate fresh MT-4 cells when the majority of the cells became infected. At 28 days after the initiation of infection, DNA was extracted from infected cells using a “rapid isolation of mammalian DNA” protocol (DNA from cells infected with the parental clone was used as a control) (29). DNA encoding CA was PCR amplified using the following primers 5′-CAA CCA TCC CTT CAG ACA GGA TCA G-3′ and 5′-TTT TGG CTA TGT GCC CTT CTT TGC C-3′, and the amplicon was sequenced directly (Beckman Coulter Genomics). The same DNA extracted from infected cells was used as a template to amplify CA DNA for insertion into pNHGcapNM, as described previously (27). Multiple clones were sequenced in this context prior to characterization of the predominant P207S, G208R, and T210K CA mutants.

Cell lysates, were resolved using 4 to 12% acrylamide gels, blotted onto nitrocellulose membranes, and probed with anti-c-Myc (9E10 hybridoma) or anti-actin (JLA20 hybridoma) antibodies (both provided by the Developmental Studies Hybridoma Bank at the University of Iowa). Membranes were then probed with fluorescently labeled goat anti-mouse secondary antibodies (Thermo Scientific) prior to scanning with a Li-Cor Odyssey scanner.

Mx2 cDNAs from the various species were inserted into the HIV-1-based expression vector, HA-CSIB (18), which expresses N-terminally hemagglutinin (HA)-tagged proteins, followed by an IRES sequence and a blasticidin resistance cassette. In addition, chimeric Mx2s (Hs₂₉Can and Can₂₉Hs) were inserted into pLKOΔ-HA-Mx2-IP, an HA-tagged derivative of pLKOΔ-MycHsMx2-IP. HOS cells were transduced with the HA-Mx2 expressing vectors, selected in antibiotic, and used as a pool for microscopy. Cells were seeded onto 24-well gelatin coated glass-bottom dishes (MatTek) and stained using anti-HA (Covance) and anti-Nup98 (Cell Signaling Technology) antibodies, followed by goat anti-mouse Alexa 488-conjugated and goat anti-rabbit Alexa 594-conjugated secondary antibodies (Molecular Probes). Cells were visualized by deconvolution microscopy as described previously (30). Image generation and colocalization analysis were completed with the SoftWorx software suite (Applied Precision). Pearson coefficient values were derived for Mx2 and Nup98 colocalization by analysis of optical sections coincident with the dorsal nuclear surface for 5 to 16 individual cells.

Analysis of CA mutants in the context of the HIV-1 capsid hexamer was done using MacPyMOL with PDB reference 3GV2 (31).

Orthologous Mx2 sequences were downloaded from complete genomes in Ensembl 73 (September 2013) and concatenated with the sequences obtained through direct sequencing. The protein sequences were aligned using Clustal Omega (32) and converted to a codon alignment using PAL2NAL (33). The best substitution model to use for phylogenetic reconstruction was determined using the Bayesian information criterion in jModeltest (34). The maximum-likelihood tree was calculated in PhyML with 1,000 bootstrap replicates (35). The codon alignment was used to detect positive selection in the primate lineage using the following Mx2 sequences: Homo sapiens (NM_002463.1), Pan troglodytes (XM_001171751.3), Pan paniscus (XM_003823906.1), Gorilla gorilla (XM_004062835.1), Pongo abelii (XM_002830701.1), Papio anubis (XM_003895463.1), Macaca mulatta (NM_001079696.1), Chlorocebus aethiops (KJ650325), Saimiri boliviensis (KJ650326), Aotus trivirgatus (KJ650327), and Macaca nemestrina (pigtail) 1 (KJ650328) and 2 (KJ650329). Positive selection was estimated using the random effect likelihood (REL) method using Hyphy made available through the Datamonkey webserver (36) and using CODEML in PAML (37). For CODEML, the codon alignment was used to construct a maximum-likelihood phylogenetic tree using RaxML v7.2.8 (38). The resulting tree was midpoint rooted and was consistent with the known phylogenetic relationships of primates. The F3x4 model for codon frequencies was used for each model, and each nested pair of models was compared using a likelihood ratio test and evaluated using the chi-squared critical value: M0 versus M3, M1a versus M2a, and M7 versus M8.

Orthologous antiviral factors frequently exhibit species-dependent activity that reflects the presence or absence of a host-pathogen molecular interaction. This interaction can occur at the interface of restriction (such as TRIM5α binding to retroviral capsids [39–42]) or antagonism (such as SIV Nef interacting with tetherin [43–46]). Accordingly, documenting species-dependent activity of restriction factors can identify sites in viral and cellular molecules that are critical for antiviral activity/antagonism.

Mx proteins are a family of genome encoded antiviral defenses whose activity has been confirmed in divergent vertebrates, including mammals (15), birds (47), and fish (48). Most mammalian Mx proteins are formed from two distinct lineages (Mx1/MxA or Mx2/MxB) that arose from an ancient duplication event. Importantly, human Mx2 and murine Mx2 are not orthologous since murine Mx1 and Mx2 are paralogous members of the Mx1 lineage with the genuine ortholog of human Mx2 having been lost in rodent and felid lineages (Fig. 1). To investigate species-dependent variation in Mx2 antiretroviral activity, we expressed human (Hs), macaque (Mac), African green monkey (AGM), ovine (Ov), and canine (Can) species variants of Mx2 in a human CD4⁺ T cell line (MT-4). Importantly, ovine and canine Mx2 are genuine orthologs of human Mx2, based on our reconstructed phylogeny of Mx genes, confirming that all of the Mx2 proteins under investigation were part of the distinct Mx2 clade (Fig. 1). We next mimicked the natural interferon induction of Mx2 expression by using a tetracycline-inducible lentiviral expression system to introduce the Mx2 variants to these cells. Doxycycline-inducible TagRFP and rhesus TRIM5α (allele Mamu1) were included as negative and positive controls, respectively, in these and subsequent experiments. To allow direct comparison of the expression levels of divergent Mx2 orthologs, all of the variants were epitope tagged. After doxycycline induction, all of the species variants of Mx2 were expressed at similar levels, as determined by Western blotting (Fig. 2A). Strikingly, Mx2 from humans, macaques and African green monkeys all inhibited HIV-1 infection by ∼10-fold, whereas ovine and canine Mx2s displayed no anti-HIV activity (Fig. 2B). Although TRIM5α was not readily visualized using Western blotting, it potently inhibited HIV-1 infection in this context (Fig. 2B, C, D, I, and J).

To further examine the spectrum of retroviruses inhibited by Mx2, we challenged our modified CD4⁺ T cells with VSVG-pseudotyped GFP-encoding retroviruses that have previously been reported to be insensitive or less sensitive to human Mx2 (feline immunodeficiency virus [FIV], equine infectious anemia virus [EIAV], and murine leukemia virus [MLV] [18, 19]). In addition, because betaretroviruses circulated in human populations in our evolutionary past, we also examined whether MPMV (49, 50) and human endogenous retrovirus K (HERV K) (51) were sensitive to Mx2. FIV (Fig. 2C), EIAV (Fig. 2D), MLV (Fig. 2E), and HERV-K (Fig. 2F) were all insensitive to inhibition by all of the Mx2 variants tested. Surprisingly, MPMV (Fig. 2G) appeared to be partially sensitive to both MacMx2 and AGMMx2 (3- to 4-fold) and was minimally sensitive to HsMx2, suggesting that Mx2 antiretroviral activity might not be limited to primate lentiviruses and that Mx2 may be capable of inhibiting retroviruses from other genera.

Human Mx2 has been shown to be active against a variety of HIV-1 strains. However, all strains tested thus far are from the M group of pandemic HIV-1 strains. We tested the HIV-1 group O molecular clone CMO2.5 (24) and found it to be as sensitive as the group M clone NL4-3 to HsMx2 (Fig. 2I and J). In contrast, AGMMx2 potently inhibited infection of the group M virus (Fig. 2I) but was nearly inactive against the group O HIV-1 strain (Fig. 2J). MacMx2, which is nearly identical to AGMMx2, was partially active against the group O HIV-1 strain. These data clearly demonstrate that an extant retrovirus, specifically HIV-1 group O, can exhibit differential sensitivity to active Mx2 proteins from different species. Thus, similar to other antiviral factors, species variants of Mx2 can possess divergent antiviral specificities and inhibit different spectra of viruses.

To investigate why the Mx2 proteins varied in their ability to inhibit retroviral infection, we first examined their subcellular localization. Deconvolution microscopic analysis revealed that MacMx2 and AGMMx2 proteins were associated with nuclei and were particularly concentrated at nuclear pores, marked by the nucleoporin Nup98 (Fig. 3A and B). In this respect, the MacMx2 and AGMMx2 proteins were indistinguishable from HsMx2. Importantly, however, the nonprimate Mx2 proteins under investigation exhibited a different pattern of localization. In the case of ovine Mx2, the protein was almost entirely cytoplasmic, and little or no ovine Mx2 protein was found in association with nuclear pores (Fig. 3A, B, and C). For canine Mx2, two distinct patterns of localization were observed. In ∼60% of cells, which appeared to express higher levels of the protein, the distribution of canine Mx2 was similar to that of primate Mx2, i.e., there was a significant degree of colocalization with nuclear pores. However, in ca. 40% of cells, which appeared to express lower levels of the protein, there was little colocalization with Nup98 (Fig. 3A, B, and C).

Inspection of the N-terminal 25 amino acids of these Mx2 proteins revealed that the canine and especially the ovine Mx2 proteins were quite different in sequence from the primate Mx2 proteins (Fig. 3D). Notably, this segment of the human Mx2 protein has been demonstrated to be required for nuclear localization (16, 52) and contains a high density of basic amino acids, a frequent characteristic of nuclear localization sequences. Moreover, we have recently shown that these N-terminal amino acids are required for the anti-HIV-1 activity of HsMx2 (18). In canine and particularly ovine Mx2 proteins, several of the basic amino acids within the N-terminal segment are divergent, perhaps explaining the reduced propensity of these proteins to associate with nuclear pores and perhaps also their lack of antiretroviral activity.

We next considered the cyclophilin dependency of Mx2 antiviral activity. Previous observations suggest that the antiviral activity of Mx2 is dependent upon an interaction between the HIV-1 capsid (CA) and a host cyclophilin (20). Notably, while Mx2 expression protected CD4⁺ SupT1 cells from HIV-1 infection, this antiviral activity was abrogated by the immunosuppressive drug cyclosporine (CsA) (20), which can disrupt cyclophilin-CA interactions. Accordingly, repeated passage of HIV-1 in the same Mx2-expressing cells selected for HIV-1 variants bearing a substitution (A88T) within the cyclophilin binding loop of CA (20). In contrast to these observations, in CD4⁺ MT-4 cells, Mx2 inhibition of HIV-1 infection remained intact in the presence of either CsA or the synthetic CsA analogue Debio-025 (D025) (53) (Fig. 4A). Under identical conditions, both CsA and D025 completely recovered lentiviral infection (SIVagmTAN) in the presence of a macaque TRIMCyp allele (25, 26, 54), confirming that both drugs were biologically active under these conditions (Fig. 4A). These data indicate that an interaction between CA and cyclophilin is not necessary for Mx2 antiviral activity in all cell types.

The observation that CsA did not rescue HIV-1 infection in the presence of Mx2 in MT-4 cells, led us to investigate whether HIV-1 could escape Mx2 via mutations at sites outside the cyclophilin binding loop, if pressured to do so in cells where CsA does not rescue infectivity. To investigate this possibility, we passaged HIV-1 in the presence of HsMx2 for 4 weeks in CD4⁺ MT-4 cells. To facilitate the quantitation of HIV-1 infection, we used a replication-competent proviral clone in which nef was replaced with enhanced green fluorescent protein (EGFP; NHG). During the first 2 weeks, replication kinetics were substantially delayed in Mx2 expressing MT-4 cells compared to noninduced controls (Fig. 4B). However, at the third passage, and at subsequent time points, the viral replication kinetics were similar, regardless of the presence of HsMx2 (Fig. 4B). We extracted DNA from the population of HsMx2-expressing infected cells at the end of the fourth passage, and the CA region of gag was PCR amplified and sequenced. Analysis of the resulting chromatogram indicated a “hot spot” of variation in the region encoding CA residues 207 to 210 (Fig. 4C). Sequencing of multiple clones indicated that the variation within the swarm was due to the presence of three separate lineages, each bearing a single CA mutation (P207S, G208R, or T210K), rather than the presence of double or triple mutants. Therefore, we generated three viral clones bearing the P207S, G208R, or T210K substitutions, in isolation. Although the parental clone was sensitive to all of the primate variants of Mx2 (Fig. 4D), the G208R and T210K variants were entirely insensitive to any of the Mx2 variants tested (Fig. 4D). Strikingly, the P207S variant behaved differently. Although this virus was partially resistant to human Mx2, this virus was fully sensitive to macaque Mx2 but entirely resistant to the nearly identical AGMMx2 (Fig. 4D). Importantly, these data indicate not only that single substitutions in CA can govern sensitivity to Mx2 activity but that divergent Mx2 variants can differentially inhibit viruses that differ by just a single CA residue. Thus, these data further suggest that, similar to other genome encoded antiviral defenses, species variants of Mx2 have divergent antiviral specificities.

The unanticipated identification of residues in the C-terminal domain of CA (Fig. 4C) that conferred resistance to Mx2-mediated antiviral activity led us to consider whether additional regions of CA might confer resistance to Mx2. We recently described a library of CA mutants that we used to probe the genetic robustness of this region of gag (27). In an attempt to identify more substitutions that might confer resistance to Mx2, we challenged CD4⁺ MT-4 T cells expressing species variants of Mx2 with 46 CA mutants from this library that produced an easily measurable infectious titer in single cycle infection assays (Fig. 5A). In accordance with our previous observations, ovine and canine variants of Mx2 did not substantially inhibit any of the mutants in the library (Fig. 5A). In contrast, while primate Mx2s inhibited the majority of HIV-1 CA mutants, seven substitutions in CA conferred resistance to one or more primate variants of Mx2 (Fig. 5A). Five of these substitutions reduced the sensitivity to all of the primate variants tested (M10I, N57S, H87R, P90T, and Q95L). In addition, two substitutions, M185I and E187V, conferred resistance to HsMx2 and AGMMx2, respectively. In order to validate the library screen, the infectivity of independent preparations of these seven mutants was quantified in cells expressing each of the species variants of Mx2 (Fig. 5C).

In agreement with our own and others' previous findings, mutations in the cyclophilin binding loop conferred resistance to Mx2. Specifically, mutations at CA residues 88 to 90, 92, and 94 have all been shown to reduce the sensitivity of HIV-1 to human Mx2 (13, 18, 20). Our screen identified a P90T substitution (Fig. 5B and C) that conferred complete resistance to all of the Mx2 variants tested. Moreover, the screen identified two additional mutants (H87R and Q95L), flanking those previously described that conferred substantial, but incomplete, resistance to Mx2. Notably, while these data highlight the importance of the cyclophilin binding loop as an Mx2 sensitivity determinant, the inability of CsA to rescue infectivity in the same cells suggests that the Mx2 resistance conferred by mutations in the cyclophilin binding loop is not solely due to the loss of an interaction with cyclophilin A (CypA).

The CA library screen also identified two other residues in the N-terminal domain of CA that promoted escape from Mx2 (Fig. 5B and C). We have previously described one of these residues (N57S) and suggested that the inability of this mutant to infect nondividing cells might underlie its reduced sensitivity to Mx2 (18, 55). Accordingly, this mutant conferred nearly complete resistance to each of the primate Mx2s. In addition, we identified a substitution in the β-hairpin close to the N terminus of CA (M10I) that also confers near complete resistance to Mx2 (Fig. 5B and C). Interestingly, like cyclophilin binding loop residues, this CA residue forms part of the cytosolic surface of the mature CA that is readily accessible to inhibitory interactions. Moreover, the same substitution, and cyclophilin binding loop residues, have previously been implicated in escape from TRIM5α (28, 56).

Outside of the CA N-terminal domain, the screen identified additional residues in the C-terminal domain of CA that influence sensitivity to Mx2 (M185I and E187V). Residue M185 is buried within intact cores and forms an important hydrophobic contact at the C-terminal domain of the CA dimerization interface (57). Although neither M185I nor E187V substitutions conferred complete resistance to all tested primate Mx2 proteins (Fig. 5C), in accordance with the initial CA library screen, M185I was substantially less sensitive to HsMx2 and the E187V mutant was almost completely resistant to AGMMx2. Interestingly, E187 is in close proximity to P207 and G208, further highlighting the importance of this region as a C-terminal Mx2 sensitivity determinant (Fig. 5B). Notably, all of the CA substitutions that result in species-specific sensitivity to primate Mx2s occur within this localized region of the C-terminal domain. Although the CA C-terminal domain is not usually implicated in mediating interactions with host factors, residues P207, G208, and T210 form the majority of a solvent accessible loop between helices 10 and 11. This sequence is exposed to the host cytosol in the context of the mature fullerene core, albeit at a position that is significantly recessed relative to the protruding CA N-terminal domains. Thus, it is possible that host factors could interact directly with this sensitivity determinant in the context of intact cores. Alternatively, it is also conceivable that these mutations exert their effect through changes in capsid morphology or stability (58) or through varying the pathway used to traverse the nuclear pore (55, 59–61). A confounding variable in these analyses is the fitness reduction observed for many of the substitutions conferring resistance to Mx2 (Fig. 4 and 5). An increase in noninfectious particles in the inoculum could, in principle, saturate Mx2 antiviral activity, leading to apparent resistance. Importantly, however, the P207S and E187V substitutions confer resistance while maintaining fitness levels similar to the parental clone. Moreover, several substitutions that also attenuate replicative fitness (L69I, N121I, R132G, and L211I [27]) did not confer resistance to Mx2.

The observation that some variants of Mx2 have no apparent antiretroviral activity, coupled with the species-dependent antiviral specificity of primate Mx2s, prompted us to assemble chimeric Mx2 molecules to identify the region(s) of Mx2 that determine antiviral activity and specificity. First, we made chimeras between human and canine Mx2s (Fig. 6A to C). Our initial human-canine chimeras suggested the importance of the N-terminal region of Mx2 (unpublished observations). This led us to make further chimeras focusing on this N-terminal region. Replacing the N-terminal 58 residues of canine Mx2 with the corresponding region of human Mx2 (Hs₅₈Can) generated a chimeric Mx2 that efficiently inhibited HIV-1 infection (Fig. 6B). In contrast, the reciprocal chimera had no anti-HIV-1 activity despite being expressed at similar levels (Fig. 6C), indicating that a major antiviral activity/specificity determinant lies within the N terminus of Mx2. To further map this determinant, we replaced the N-terminal 29 amino acids of CanMx2 with HsMx2 (Hs₂₉Can). As before, this chimeric Mx2 efficiently inhibited HIV-1 infection, whereas the reciprocal chimera had no anti-HIV-1 activity (Fig. 6B), despite similar levels of expression (Fig. 6C). The inhibitory action of these chimeras indicates that a major antiviral activity/specificity determinant resides within the first 29 amino acids of Mx2. Interestingly, the N-terminal region of Mx2 is the most divergent section of mammalian Mx2 proteins. Moreover, it contains the determinant that is required for nuclear localization, which is more robust in the case of the primate Mx2 proteins than for canine Mx2 (Fig. 3).

Because the N-terminal region of human Mx2 is required for correct nuclear localization (16, 52), we reasoned that replacing the first 29 residues of canine Mx2 with those of human Mx2 would generate a chimeric Mx2 whose localization was reminiscent of primate Mx2. Indeed, deconvolution microscopic analysis revealed that the Hs₂₉Can chimera was associated with nuclei and was particularly concentrated at nuclear pores, marked by the nucleoporin Nup98 (Fig. 6D and E). In this respect, the chimeric protein was indistinguishable from HsMx2. In contrast, the reciprocal chimera exhibited a different pattern of localization. In the presence of the canine N terminus, two distinct patterns of localization were observed. Similar to canine Mx2, in ∼60% of cells there was a significant degree of colocalization with nuclear pores. However, in ca. 40% of cells, there was little colocalization with Nup98 (Fig. 6D and E). The correlation between nuclear pore localization and anti-HIV-1 activity suggests that localization to the nuclear pore could be a requirement for anti-HIV-1 activity.

To further investigate Mx2 antiviral specificity and remove the variable of differential localization that might confound an analysis of CanMx2/HsMx2 chimeras, we made chimeras between the closely related (99% identical), nuclear-pore-localized, macaque and AGM Mx2 proteins (Fig. 6F to I). Of these two Mx2 proteins, only macaque Mx2 inhibits the P207S HIV-1 CA mutant (Fig. 4D). Therefore, we could exploit this observation to uncover which amino acids define the divergent specificities of these nearly identical factors. In accordance with the human-canine chimeras, replacing the N-terminal 333 amino acids of AGMMx2 with the corresponding region of MacMx2 generated a chimeric Mx2 (Mac₃₃₃AGM) that efficiently inhibited infection of the P207S CA mutant. In contrast, the reciprocal chimera had no activity against the P207S virus but retained activity against the parental HIV-1 clone (Fig. 6G). To further map this specificity determinant, we replaced the N-terminal 90 amino acids of AGMMx2 with the same region of MacMx2 (Mac₉₀AGM). Because MacMx2 and AGMMx2 are so closely related, this chimera differs at just three positions from AGMMx2 (Fig. 6I), and these three changes are concentrated in an 8-amino-acid stretch (residues 37 to 44). Remarkably, substituting these three residues (Mac₉₀AGM) completely transferred antiviral specificity and generated a chimeric Mx2 that efficiently inhibited infection by the HIV-1 P207S mutant (Fig. 6G). In contrast, the reciprocal chimera (AGM₉₀MAC) had no activity against this mutant but retained the ability to inhibit infection with the parental HIV-1 strain. The observation that all AGM and macaque chimeras were able to inhibit the parental HIV-1 clone indicates that these chimeras were all active antiviral factors capable of inhibiting HIV-1 infection. Thus, amino acids 37 to 44 must constitute an antiviral specificity determinant within these functional Mx2 proteins. As before, Mx2 antiviral activity was also highly specific, and no inhibition of MLV infection was observed in the presence of any chimeric Mx2s. Considered together, these data suggest that a major antiviral specificity determinant of Mx2 resides in a localized 8-amino-acid segment close to the N terminus of Mx2.

In light of the observation that this 8-amino-acid region can govern Mx2 antiviral specificity for the HIV-1 P207S CA mutant, we reasoned that these same residues could define antiviral specificity for group O HIV-1. To investigate this possibility, we examined the ability of human and AGM chimeric Mx2s to inhibit group O HIV-1. Human and AGM Mx2 only differ at positions 37 and 39 within this 8-amino-acid region (Fig. 6J) so we constructed human and AGM Mx2s that were chimeric only at these two positions. Similar to human and AGM Mx2 variants, both chimeras were active against group M HIV-1 (NL4-3). Strikingly, despite being active against group M HIV-1, human Mx2 with AGM residues 37 and 39 (HsRTA) was unable to inhibit group O HIV-1 (Fig. 6K). Conversely, humanizing residues 37 and 39 of AGM Mx2 (AGMGTV) conferred potent anti-group O activity to this previously inactive variant (Fig. 6K). Importantly, these data highlight that residues within this localized antiviral specificity determinant can govern activity against HIV-1 derived from extant viruses. Thus, this region likely governs specificity for circulating viruses and is not limited to determining activity against HIV-1 CA mutants.

The observation that primate Mx2s exhibit species-dependent antiviral activity raised the possibility that Mx2 and primate lentiviruses have been involved in a molecular “arms race” during primate evolution. Therefore, we investigated whether signatures of positive selection exist in primate Mx2 sequences. We confined this analysis to primate sequences since thus far we have only observed Mx2 antiviral activity against primate viruses. We subjected sequences from NCBI databases or generated as part of the present study to positive selection analysis. Using our data set, we calculated an average ratio of nonsynonymous to synonymous evolutionary changes (dN/dS) of 0.44 using CODEML, a finding consistent with purifying selection having been the dominant selective force during Mx2 evolution. However, both the REL and CODEML methods identified codons with high dN/dS ratios (Fig. 7A). The nested models from CODEML were compared, and the M3 model gave a significantly better fit than the M0 model, suggesting that distinct selective pressures have acted on specific sites during primate evolution. The REL and M3 methods both identified a set of 10 positively selected sites with Bayes factors above 50 and posterior probabilities above 0.95 (listed in Table 1). Interestingly, two positively selected sites (residues 37 and 44) lie within the 8-amino-acid specificity determinant identified using chimeric macaque and AGM Mx2s. One of these sites, residue 37, had a P value of 0.998 and the highest Bayes factor within the N-terminal region (95.08), suggesting this residue could have been of particular importance during primate evolution. We therefore examined the ability of this residue to govern Mx2 antiviral specificity. Remarkably, making a single substitution at residue 37 was sufficient to alter Mx2 antiviral specificity. Specifically, replacing AGM Mx2 residue 37 with the corresponding macaque residue (AGMR37G) was sufficient to confer potent antiviral activity to AGMMx2 against the P207S HIV-1 CA mutant (Fig. 7D). Conversely, making the reciprocal substitution in macaque Mx2 (MacG37R) ablated antiviral activity against the P207S mutant (Fig. 7D). Crucially, both point mutants were expressed at similar levels (Fig. 7C) and were active against the parental HIV-1 clone (Fig. 7D). Thus, in this instance, amino acid 37 governs species-specific Mx2 antiviral activity against the HIV-1 P207S mutant.