AXL-dependent infection of human fetal endothelial cells distinguishes Zika virus from other pathogenic flaviviruses

Significance

Zika virus (ZIKV) causes microcephaly, whereas other related pathogenic flaviviruses do not. To reach the fetal brain, a virus must be transported from the maternal to the fetal circulation, which requires crossing of the placental barrier. Our studies demonstrate that mammalian cell-derived ZIKV, but not two other globally relevant flaviviruses, efficiently infects fetal endothelial cells, a key component of the placental barrier, because only ZIKV can efficiently use the cell-surface receptor AXL. These data suggest that use of AXL allows ZIKV to enter the fetal bloodstream to gain access to other fetal tissues. Thus, this study provides insight into the unique properties of ZIKV that contribute to its ability to cause microcephaly and other congenital infections and diseases.

Abstract

Although a causal relationship between Zika virus (ZIKV) and microcephaly has been established, it remains unclear why ZIKV, but not other pathogenic flaviviruses, causes congenital defects. Here we show that when viruses are produced in mammalian cells, ZIKV, but not the closely related dengue virus (DENV) or West Nile virus (WNV), can efficiently infect key placental barrier cells that directly contact the fetal bloodstream. We show that AXL, a receptor tyrosine kinase, is the primary ZIKV entry cofactor on human umbilical vein endothelial cells (HUVECs), and that ZIKV uses AXL with much greater efficiency than does DENV or WNV. Consistent with this observation, only ZIKV, but not WNV or DENV, bound the AXL ligand Gas6. In comparison, when DENV and WNV were produced in insect cells, they also infected HUVECs in an AXL-dependent manner. Our data suggest that ZIKV, when produced from mammalian cells, infects fetal endothelial cells much more efficiently than other pathogenic flaviviruses because it binds Gas6 more avidly, which in turn facilitates its interaction with AXL.

Zika (ZIKV), West Nile (WNV), and dengue (DENV) viruses are closely related, and belong to the Flavivirus genus in the Flaviviridae family. Although a causal relation between ZIKV and microcephaly has been established by human and animal studies (1–7), it remains unclear why only ZIKV, but not other pathogenic flaviviruses, causes congenital diseases. Although WNV is known to infect neuronal cells and results in encephalitis (8), it does not cause microcephaly. DENV is not generally neurotropic and is not linked to congenital defects.

To reach the fetal brain, a virus must be transported from the maternal to the fetal circulation, which necessitates crossing of the placental barrier. In the placenta, fetal blood in capillaries is separated from maternal blood by placental barrier cells, namely trophoblasts and fetal endothelial cells. Recent studies indicate that the placenta and its barrier cells are infected by ZIKV, and fetal brain lesions develop in mice, pigtail macaques, and humans (1–6, 9). However, it remains unclear why only ZIKV, and not other neurotropic flaviviruses, results in microcephaly and other congenital disorders.

Although bona fide entry receptors for flaviviruses remain unknown, many cell surface-expressed molecules contribute to infection, including C-type lectins dendritic cell-specific intercellular adhesion molecule-3-grabbing nonintegrin (DC-SIGN) and DC-SIGN–related protein (L-SIGN) (10, 11) and phosphatidylserine (PS) receptors (12–15)_. PS receptors, which serve as entry cofactors for flaviviruses, include members of the TIM (T-cell Ig mucin) family and the TAM (TYRO3, AXL, and MERTK) family. TIM-family receptors bind PS directly (14, 15), whereas TAM-family members bind PS indirectly, through the soluble intermediates Gas6 (growth arrest-specific 6) and protein S present in serum and other bodily fluids (16, 17). Whereas Gas6 binds to all three TAM family members with high affinity, protein S binds to TYRO3 and MERTK, but not to AXL (17). The TAM receptor AXL was recently shown to support ZIKV infection of human foreskin fibroblasts (12), and its expression was noted in the brain and neuroprogenitor cells (18–21). However, its deletion had no effect on ZIKV infection of induced pluripotent human stem cell-derived neuroprogenitor cells or cerebral organoids (22) or on virus accumulation of the eye, brain, or testis in Axl^−/− mice (23, 24). Expression of other flavivirus entry cofactors (genatlas.medecine.univ-paris5.fr/fiche.php?onglet=4&n=26146, genatlas.medecine.univ-paris5.fr/fiche.php?onglet=4&n=1364, and refs. 22 and 25) in addition to AXL, however, might have compensated for the absence of AXL in these cells and tissues.

In contrast, we show here that AXL is the only relevant ZIKV entry cofactor expressed on fetal endothelial cells, and that when produced in mammalian cells, only ZIKV, but not WNV or DENV, can use AXL, because it more efficiently binds Gas6. These differences may help explain why only ZIKV, and not other flaviviruses, can access the fetal bloodstream to infect fetal tissues and cause microcephaly.

Results

Human Umbilical Vein Endothelial Cells Are More Susceptible to ZIKV than to DENV or WNV.

To investigate why ZIKV, and not closely related flaviviruses (e.g., WNV and DENV), causes microcephaly, we initially evaluated the susceptibility to ZIKV, DENV, and WNV of fetal endothelial cells, which serve as a barrier between the placenta and fetal tissues. We propagated viruses in a mammalian cell line (Vero 76), rather than a commonly used insect cell line, to more accurately reflect the virus population generated during maternal infection. Human umbilical vein endothelial cells (HUVECs) from three independent donors (donors 1 to 3) or pooled HUVECs from additional donors (donor 4) were infected at various multiplicities of infection (MOIs of 0.1 to 1.0) with ZIKV (FSS13025, Cambodia), DENV (serotype 2, New Guinea C), or WNV (lineage I, New York 1999). At all MOIs and in all four HUVEC preparations, ZIKV infection was much more efficient than WNV or DENV (Fig. 1 A–C and Fig. S1 A–C). To confirm that the higher level of ZIKV infection was not an artifact of preferential recognition of ZIKV E protein by the pan-flavivirus antibody 4G2, we compared 4G2 staining with that by virus-specific antibodies and observed no qualitative difference (Fig. S1D and Table S1). ZIKV progeny viruses in the culture supernatants, as measured by plaque assays, also showed ∼100- and 1,000-fold higher titers than those from WNV- and DENV-infected cells, respectively (Fig. 1D). Thus, ZIKV infects primary fetal endothelial cells to a substantially greater level than does DENV or WNV.

Fig. 1.

HUVECs are more susceptible to ZIKV infection than to DENV or WNV. (A) HUVECs were infected with Vero 76-produced ZIKV, DENV, or WNV at an MOI of 1. Infection levels, assessed at 24 h postinfection by staining permeabilized cells with the pan-flavivirus antibody 4G2, were normalized to that of ZIKV within each donor. Averages ± SD of three experiments performed in duplicate are shown. ***P < 0.0001. See also Fig. S1. (B) Infection profiles of a representative experiment from A are shown as histograms, where infected cells (colored lines) are compared with mock-infected cells (gray lines). (C) Similar to A, except that cells were fixed and stained with the antibody 4G2 on multiwell plates. (D) The progeny viruses in the supernatants from the experiments in A were quantified by plaque assays in Vero cells. The average titers based on three independent experiments are presented as plaque-forming units per milliliter.

AXL Is the Primary ZIKV Entry Cofactor in HUVECs.

To identify the mechanism by which ZIKV infected HUVECs, we measured the surface expression of several established flavivirus entry cofactors: TIM- and TAM-family receptors, and DC-SIGN and L-SIGN. All HUVECs expressed AXL, as previously reported (26), and, at lower levels, MERTK (Fig. 2A and Fig. S2A). No expression was observed for other entry cofactors, indicating they do not contribute to the infection of HUVECs by these viruses. The specificity of the antibodies was confirmed by staining relevant molecules ectopically expressed in HEK293T cells (Fig. S2B).

Fig. 2.

AXL is the primary ZIKV entry factor in HUVECs. (A) The cell-surface expression of the indicated proteins (red) or isotypes (gray) was assessed in HUVECs. See also Fig. S2. (B) HUVECs were preincubated with an anti-AXL antibody (AF154) or control IgG, and infected with ZIKV or IAV. Infection levels were normalized to those of cells infected without antibody. (C, Left) Similar to A, except that an anti-MERTK antibody (AF891) was compared with the anti-AXL antibody. Infection levels were normalized as in A. (C, Right) The ability of the antibody to bind MERTK on HUVECs is shown. (D, Left) The efficiency of AXL gene editing via the CRISPR/Cas9 method, directed by an AXL-specific sgRNA in HUVECs (AXL^KO cells; red) and an untargeted sgRNA (control cells; blue), is shown. Cells were stained with anti-AXL antibody (clone 108724). (D, Right) These cells were infected with ZIKV or IAV and infection levels were normalized to those of control cells for each virus. (E, Left) AXL expression was analyzed, using the anti-AXL antibody (clone 108724), in HUVECs transfected with the indicated siRNA. (E, Right) These cells were infected with ZIKV at an MOI of 1, and infection levels were normalized to those of cells transfected without any siRNA. ZIKV was produced in Vero 76 cells and IAV in Madin–Darby canine kidney (MDCK) cells. (B–E) Averages ± SD of three (B and C) or five (D and E) experiments performed in duplicate are shown. **P < 0.001, ***P < 0.0001.

To determine whether AXL expression on HUVECs contributed to ZIKV infection, cells were preincubated with an anti-AXL antibody, which blocks Gas6 binding but induces AXL phosphorylation (27, 28). ZIKV infection in HUVECs was effectively inhibited by the anti-AXL antibody, but not by control IgG (Fig. 2B). The same antibody did not inhibit the infection of influenza A virus (IAV), which does not use any TIM or TAM family members (13, 15). We similarly assessed the role of MERTK in ZIKV infection of HUVECs. An anti-MERTK antibody, which also blocks Gas6 binding (29), did not inhibit ZIKV infection (Fig. 2C). To corroborate these findings, we used two genetic approaches: CRISPR/Cas9 gene editing and siRNA gene silencing. Both methods reduced AXL expression to undetectable levels, and abolished ZIKV infection as judged by intracellular staining of E protein (Fig. 2 D and E). Collectively, these data indicate that AXL is the primary cofactor for ZIKV infection of fetal endothelial cells.

ZIKV, but Not DENV or WNV, Efficiently Uses AXL.

To investigate whether the differential ability of ZIKV, WNV, and DENV to infect HUVECs was due to distinct AXL-use patterns, we evaluated the effect of the AXL antibody on infection of HUVECs by these viruses produced from Vero 76 cells. We used a high MOI of 20 to obtain measurable intracellular staining levels of WNV and DENV infection (Fig. S3A) and an MOI of 1 for progeny virus titering (Fig. 3A). Infection of WNV and DENV was not affected by the anti-AXL antibody, whereas ZIKV infection was markedly reduced. We then conducted similar experiments in Vero 76 cells, which support efficient infection of all three viruses, likely because multiple flavivirus entry cofactors are expressed on these cells. We obtained similar results from these cells, using reporter virus-like particles (RVPs), which are capable of single-round infection (Fig. S3B). Again, only ZIKV RVP infection was inhibited by the anti-AXL antibody. We then assessed AXL use by these viruses by titering their progeny viruses produced from AXL-KO and control HUVECs (Fig. 3B). Whereas ZIKV titer was substantially decreased, WNV and DENV titers were only marginally reduced. We confirmed these AXL-use patterns using HEK293T cells ectopically expressing AXL. Note that parental HEK293T cells lack endogenous expression of DC-SIGN and AXL, and DC-SIGN was included as a control, because it is used by many flaviviruses to enter cells (10, 11). Only ZIKV infection was enhanced in AXL-expressing cells, whereas infection by all three flaviviruses was increased in DC-SIGN–expressing cells, compared with the parental HEK293T cells (Fig. 3C). Although the basal infection of the parental HEK293T cells by WNV and DENV is much higher than that by ZIKV, AXL use by WNV or DENV was not apparent even at low MOIs, in contrast to that seen with DC-SIGN. These data verify that the low level of WNV and DENV infection observed in HUVECs was not mediated by AXL.

Fig. 3.

ZIKV, but not DENV or WNV, uses AXL efficiently. (A and B) HUVECs, preincubated with 50 nM anti-AXL antibody (AF154) or control IgG (A), or HUVEC AXL^KO and control cells (B), were infected with Vero 76-produced ZIKV, DENV, or WNV at an MOI of 1. The progeny viruses at 24 h postinfection were quantified by plaque assays in Vero cells. Results are expressed as plaque-forming units per milliliter. (C, Left) Expression levels of AXL or DC-SIGN in transduced HEK293T cells are shown. (C, Right) Parental HEK293T-, AXL-, or DC-SIGN–transduced cells were infected with ZIKV, DENV, or WNV at the indicated MOI. Results are presented as percent infected cells. Averages ± SD of three (A and C) or five (B) experiments performed in duplicate are shown. *P < 0.01, **P < 0.001, ***P < 0.0001.

Gas6 Binds to ZIKV, but Not to DENV or WNV.

To assess whether AXL utilization by ZIKV was mediated by the AXL ligand Gas6, we first attempted to inhibit virus–AXL interaction with C-Gas6-Ig, an Ig-fusion form of a human Gas6 variant that contains only the C-terminal half (AXL-binding domain) of the molecule. This construct thus lacks the PS-binding domain. TIM1(AA)-Ig does not bind PS (13), and was used as a negative control. At 0.4 μg/mL, C-Gas6-Ig substantially inhibited the infection of HUVECs by ZIKV (Fig. 4 A and B), indicating that ZIKV associates with AXL at the same site where Gas6 binds, and suggesting that AXL use by ZIKV is mediated by Gas6. Note that serum concentrations of Gas6 in healthy humans range from 13 to 100 ng/mL (30–32), and that Gas6 in medium containing 10% (vol/vol) FBS is sufficient for maximal transduction by various viruses mediated by AXL (33). Although various tissues and cells produce Gas6, the scale of this production is unknown.

Fig. 4.

Gas6 binds to ZIKV but not to DENV or WNV. (A) HUVECs were preincubated with C-Gas6-Ig or TIM1(AA)-Ig, and infected with Vero 76 cell-produced ZIKV, DENV, or WNV at an MOI of 1. Infection levels were normalized to those of cells infected in the absence of any Ig-fusion protein within each virus. (B) Progeny viruses in the culture supernatants from the experiments in A were quantified by plaque assays in Vero cells. Results are expressed as plaque-forming units per milliliter. (C and D) ZIKV, DENV, or WNV, produced in Vero 76 cells, was incubated with Gas6-Ig, C-Gas6-Ig, or TIM1-Ig and immunoprecipitated using protein A-Sepharose beads. (C) The RNA of the bound viruses was extracted and quantified by RT-qPCR. Binding is represented as fold increases normalized within each virus to that of virus incubated with C-Gas6-Ig. See also Fig. S6. (A–C) Averages ± SD of three experiments performed in duplicate (A and B) or quadruplicate (C) are shown. ***P < 0.0001. (D) Bound viruses were analyzed by Western blot (WB) using antibodies against E protein. A fraction (15%) of the input virus was loaded as a quantity control. A representative experiment of three performed is shown. IP, immunoprecipitation.

To investigate whether ZIKV, but not WNV or DENV, interacts with Gas6, we performed a modified immunoprecipitation assay. Viruses were preincubated with Gas6-Ig, C-Gas6-Ig, or TIM1-Ig, and bound to protein A-Sepharose beads. Captured viruses were quantified by RT-quantitative (q)PCR of the viral RNA or visualized by Western blot (Fig. 4 C and D and Fig. S4). C-Gas6-Ig was used as a negative control, and TIM1-Ig was used as a positive control, because all three viruses use TIM1 to infect cells (Fig. S5) (13, 15). ZIKV, but not DENV or WNV, was captured by Gas6-Ig immobilized on protein A beads. As expected, no virus was captured by C-Gas6-Ig, and all three viruses were captured by TIM1-Ig. Together, these data demonstrate that ZIKV, but not WNV or DENV, can efficiently use AXL, because only ZIKV is able to bind Gas6 efficiently.

Insect Cell-Derived DENV and WNV also Use AXL.

Although our data indicate that Vero 76-produced DENV and WNV do not use AXL, AXL-dependent infection by DENV and WNV has been reported by others (14, 34). We investigated whether differences in virus producer cells could influence infection outcomes, and therefore repeated our infection studies with virus produced in C6/36 insect cells. Note that viruses were produced in insect cells in the studies by Meertens et al. (14) (DENV) and Bhattacharyya et al. (34) (WNV) but in Vero 76 cells in our study. Infection of HUVECs by C6/36-produced DENV and WNV was much more efficient than that produced in Vero 76 cells, when infected at the same MOI, and was efficiently inhibited by the anti-AXL antibody, judged by both intracellular staining and plaque assays of progeny viruses (Fig. 5 A and B). Moreover, infection of C6/36-produced WNV and DENV was substantially reduced in AXL^KO HUVECs relative to the parental HUVECs (Fig. 5 C and D). To verify that mammalian cell-derived WNV and DENV do not use AXL, we produced these viruses in human cell lines A549 and Huh7. As with Vero 76-produced virus, ZIKV derived from these cell lines efficiently used AXL but DENV and WNV did not (Fig. S6). Thus, DENV and WNV can use AXL to infect HUVECs when produced from insect cells but not when produced from mammalian cells.

Fig. 5.

DENV and WNV produced in C6/36 cells use AXL, whereas those produced in Vero 76 cells do not. HUVECs preincubated with 50 nM anti-AXL (AF154) or control antibody (A and B), or HUVEC AXL^KO or control cells (C and D), were infected at an MOI of 1 with DENV or WNV produced either in Vero 76 or in C6/36 cells. (A and C) Results are presented as percent infected HUVECs. (B and D) Progeny viruses at 24 h postinfection were quantified by plaque assays in Vero cells. Results are expressed as plaque-forming units per milliliter. Averages ± SD of three experiments performed in duplicate are shown. **P < 0.001, ***P < 0.0001.

Discussion

Our current studies show that mammalian cell-derived ZIKV efficiently infects one of the two major placental barrier cells, namely fetal endothelial cells, whereas WNV and DENV do not under the same experimental conditions. This implies that fetal endothelial cells serve as a barrier to WNV and DENV, but not to ZIKV, and may contribute to the ability of ZIKV to disseminate to other fetal tissues. Our studies also show that AXL is the primary cofactor for ZIKV infection in these cells, and suggest that the differential infection of HUVECs by ZIKV, WNV, and DENV is due in large part to the different efficiencies with which they bind Gas6 and use AXL.

Tabata et al. recently reported that trophoblasts, the first layer of the placental barrier, expressed TIM1 and were infected by ZIKV (9). Although it is not yet established, WNV and DENV likely infect primary trophoblasts because they both efficiently use TIM1 (Fig. S5) (13, 15). This implies that the trophoblast layer might not be an effective barrier for WNV, DENV, or ZIKV. In comparison, our studies show that only ZIKV efficiently infects fetal endothelial cells, another key placental barrier cell. A recent study by Miner et al. showed that both trophoblasts and fetal capillaries were infected and injured by ZIKV infection in a pregnancy model in mice (3). Accordingly, it will be of interest to determine whether fetal endothelial cells in the mouse model present a barrier to DENV and WNV as suggested by our studies.

TAM receptors mediate phagocytosis of apoptotic cells in various tissues, including blood–brain barrier endothelial cells, retinal pigment epithelial cells of the eye, and Sertoli cells of the testes (25, 35, 36). This expression pattern is consistent with ZIKV pathogenesis observed in the brain and eye and its ability to transmit sexually (4, 37, 38). However, recent studies did not observe differences in ZIKV infection of the eye, brain, or testes between wild-type and Axl^−/− mice (23, 24). Similarly, Wells et al. observed no difference in ZIKV infection between wild-type and AXL-KO stem cell-derived neuroprogenitor cells and organoids (22). Of note, Axl^−/− mice exhibited elevated blood–brain barrier permeability (39, 40). In addition, TIM1 is expressed at high levels in retinal pigment epithelium and uroepithelium (genatlas.medecine.univ-paris5.fr/fiche.php?onglet=4&n=26146), DC-SIGN and L-SIGN are expressed in brain microvascular cells (25), and TYRO3 is expressed in the brain and neuroprogenitor cells (genatlas.medecine.univ-paris5.fr/fiche.php?onglet=4&n=1364 and ref. 22). Thus, it remains possible that increased endothelium permeability in Axl^−/− mice could have contributed to the observed outcomes, and the absence of AXL in those tissues of Axl^−/− mice was compensated for by other entry cofactors. In contrast, no functional flavivirus entry cofactor other than AXL is expressed in HUVECs (Fig. 2A), explaining its greater contribution to ZIKV infection in these cells than in those tissues.

One outstanding question in the field is how PS receptors promote flavivirus infection, because structural studies show that the E-protein shell occludes most of the virion membrane (41–43). However, flavivirus particles assume many asymmetric states (44–46) and are in continuous dynamic motion (47, 48), which likely exposes patches of the virion membrane. In addition, mosaic virions—those that are both mature and immature in patches—have been observed functionally and by cryo-EM (47, 49–52). At present, it is unclear why only ZIKV preferentially binds Gas6 whereas all three viruses bind TIM1 (Fig. 4 and Fig. S4). One possible explanation is that because Gas6 is a larger protein, relative to TIM1, it requires larger patches of exposed virion membrane, which might be more available on ZIKV than on WNV or DENV. Greater membrane exposure on ZIKV could be achieved if its structural proteins facilitate increased dynamic motion, or if more ZIKV particles are in a mosaic state. It also remains uncertain why insect cell-derived DENV and WNV, but not those produced in mammalian cells, can use AXL. Different lipid composition in insect cells compared with mammalian cells and the lower temperature used for virus production (e.g., 28 °C) could affect assembly and thus the ensembles of conformational states adopted. Clearly, the detailed mechanism by which ZIKV binds Gas6 and uses AXL warrants further investigation.

In summary, our data suggest that the unique ability of ZIKV to infect fetuses and cause congenital malformations may derive, at least in part, from its capacity to bind Gas6 and use AXL, which may enable it to infect fetal endothelial cells and cross the placental barrier more efficiently than other flaviviruses.

Materials and Methods

See SI Materials and Methods for detailed methods.

Virus Infection and Inhibition Assays.

For infection assays, cells were incubated at 37 °C with viruses at the indicated MOI for 6 h (HUVECs) or 1 h (HEK293T and Vero 76 cells), and then further grown in fresh medium for 24 h (HUVECs and Vero 76 cells) or 48 h (HEK293T cells). Cells were stained with the pan anti-flavivirus antibody 4G2, unless otherwise stated, or with the anti-IAV antibody (clone C179; Takara), and analyzed by flow cytometry. For inhibition assays, the indicated cells were preincubated with an anti-AXL (AF154; R&D Systems) or anti-MERTK (AF891; R&D Systems) antibody for 20 min at room temperature, infected with replication-competent viruses or reporter virus-like particles, and analyzed as described above.

CRISPR/Cas9- and siRNA-Mediated Silencing of AXL Expression.

To generate the AXL gene-edited HUVECs, an AXL-specific single-guide (sg)RNA was cloned into the lentiCRISPR v2 plasmid (Addgene; 52961). An untargeted sgRNA was used as a control. HUVECs were transduced with lentiviruses coexpressing Cas9 and sgRNA, and selected with 1.5 μg/mL puromycin at 24 h posttransduction for 4 d. To silence AXL expression, HUVECs (at 80 to 85% confluence) were transfected with 25 nM untargeted or AXL-specific siRNA using DharmaFECT 4 reagent (Dharmacon). The day after transfection, cells were detached and plated for further infection assays and the analysis of AXL cell-surface expression.

Gas6-Ig Binding to Viruses.

ZIKV, DENV, and WNV were quantified by RT-qPCR of their respective NS3 gene, mixed with Gas6-Ig, C-Gas6-Ig, or TIM1-Ig, and incubated at 37 °C with 5% CO₂. After 1 h, protein A-Sepharose beads, preblocked with 2% (vol/vol) BSA, were added and further incubated for 1 h at room temperature on a rocking platform. After washing, bound viruses were quantified by RT-qPCR of viral RNA or visualized by Western blot analyses using anti-E protein antibodies 4G2 (ZIKV and DENV) or 3.91D (WNV).

Data and Statistics.

The difference between groups was tested using an unpaired two-tailed t test with Welch’s correction. The null hypothesis was rejected when P < 0.01.