Innate immune control of influenza virus interspecies adaptation via IFITM3

IFITM3 deficiency lowers the minimum infectious dose of avian influenza viruses in mice

Intrinsic immunity is presumed to be among the factors responsible for preventing infection when virus inoculums are below a minimum infectious dose threshold. However, this fundamental concept lacks concrete experimental evidence. IFITM3 limits the severity of influenza virus infections in humans and mice^{16,17,21,22,23}, but whether it influences the minimum viral dose required for productive infection in vivo has not been tested. We infected WT and Ifitm3^−/− mice with doses of 1, 10, or 50 TCID50 of H5N1 avian influenza virus and measured lung viral loads at day 3 post-infection. Infection with 1 TCID50 of H5N1 influenza virus resulted in significant viral replication in the lungs of Ifitm3^−/− mice, approaching titers of 10⁴ TCID50/mL, while live virus was not detected in the lungs of WT mice (Fig. 1a). Doses of 10 and 50 TCID50 caused detectable viral loads in both WT and KO mice, with KO mice showing significantly higher viral titers at both doses (Fig. 1a). Examination of lung IL-6 and IFNβ as measures of inflammation induced by viral replication substantiated these titer results. Both lung IL-6 and IFNβ were nearly undetectable in WT mice infected with 1 TCID50 but were significantly above baseline in all other samples, including Ifitm3^−/− mice infected with only 1 TCID50 (Fig. 1b, c). We next infected WT and KO mice with a second avian virus using doses of 1 and 10 TCID50. We again observed that 1 TCID50 of the H7N3 virus was sufficient to infect Ifitm3^−/− mice as indicated by robust replication and induction of IL-6 and IFNβ, whereas the same viral dose did not productively infect WT mice (Fig. 1d–f). A dosage of 10 TCID50 resulted in productive virus replication in both WT and Ifitm3^−/− mice, though viral titers and cytokine induction were each higher in the KO animals (Fig. 1d–f).

WT and Ifitm3^−/− mice were intranasally infected with (a–c) 1, 10, or 50 TCID50 of H5N1 avian influenza strain (2 independent experiments for doses 1 and 10 (n = 10 mice) and 1 experiment for dose of 50 (n = 5 mice)) or with (d–f) 1 or 10 TCID50 of H7N3 avian influenza strain (n = 5 mice). a, d Viral titers from lung homogenates at day 3 post infection. b, e ELISA quantification of IL-6 levels in lung homogenates at day 3 post infection. c, f ELISA quantification of IFNβ levels in lung homogenates at day 3 post infection. All error bars represent SEM. Comparisons were analyzed by one-way ANOVA followed by Tukey’s multiple comparisons test. Only comparisons between WT and Ifitm3^−/− mice for each dose are shown. a–f Each data point represents an individual mouse. The numbers shown above the graph represent exact p-values. Source data are provided as a Source Data file.

IFITM3 deficiency lowers the minimum infectious dose of avian influenza viruses in human cells

To extend our conclusions to a human system, we tested a broad panel of influenza viruses isolated from animals in a relevant human cell type, A549 lung epithelial cells. For our study, we compared infection rates of 11 avian, 3 swine, and 2 human influenza viruses representing distinct and diverse hemagglutinin protein (HA) subtypes (strain details in Supplementary Table 1 and HA-based phylogenetic tree in Supplementary Fig. 1). Notably, 2 of the swine viruses are known to have infected humans making them confirmed zoonotic viruses^29,30. We first examined single-cycle infections of these viruses in the absence of trypsin in control A549 cells versus those in which IFITM3 was depleted by shRNA knockdown (Fig. 2a). The gating strategy for determining infected A549 cells is included in Supplementary Fig. 2a. The IFITM3-knockdown cells were universally infected at a higher rate (Fig. 2b, c, e and Supplementary Fig. 3), demonstrating that each of these diverse viruses was restricted by IFITM3 in human cells. Furthermore, IFITM3-knockdown cells maintained higher infection susceptibility than control cells after treatment with type I interferon (Fig. 2b, c, e and Supplementary Fig. 3), demonstrating that IFITM3 plays a critical and non-redundant role in the human antiviral interferon response against diverse influenza viruses. Since macrophages are also a relevant target of influenza virus³¹, we examined infection of differentiated human THP-1 cells. Once again, infection by the human-, avian-, and swine-origin viruses was significantly potentiated in the human macrophages lacking IFITM3 with or without interferon treatment (Fig. 2b, d, e; Supplementary Fig. 2b and Supplementary Fig. 4). Increased infection was also observed for IFITM3 KO HAP1 fibroblasts (Supplementary Fig. 2c and Supplementary Fig. 5) and for IFITM1/2/3 KO HeLa cells (Supplementary Fig. 2d, Supplementary Fig. 6). Of note, we determined sialic acid receptor binding preferences for viruses we tested and found that the human- and swine-origin viruses showed a preference for a-2,6-linked sialic acid while many of the avian viruses showed a preference for a-2,3-linked sialic acid (Supplementary Table 1). Despite these differences in receptor binding preference, all the viruses we examined were able to infect every cultured human cell type that we tested (Fig. 2b, e and Supplementary Figs. 2–6). Overall, these data demonstrate that IFITM3 deficiency broadly increases infection of human cells with influenza viruses of animal origin.

a Schematic of in vitro infection with animal-origin influenza viruses and representative example flow cytometry dot plots from infected A549 human lung cells. b, e The indicated A549 cells or THP-1 differentiated macrophages were treated +/- IFNβ for 18 h followed by infection with the indicated viruses (MOI 1) for 24 h. Percent infection was determined by flow cytometry and normalized to respective control cells without IFNβ pre-treatment. Error bars represent SEM. Only statistical comparisons between shControl versus shIFITM3 and WT versus IFITM3^−/− are shown, determined by one-way ANOVA followed by Tukey’s multiple comparisons test. Data are representative of 3 independent experiments, each performed in triplicate (n = 9). c, d Western blots of cell lysates at 18 h +/− IFNβ treatment. Note that commercial IFITM3 antibodies weakly detect IFITM2 in addition to IFITM3. The numbers shown above the graph represent exact p values. Source data are provided as a Source Data file. a Created in BioRender. Denz, P. (2024) BioRender.com/z17v580. Gating strategies for flow cytometry are depicted in Supplementary Fig. 2a. A549 cells and Supplementary Fig. 2b. THP-1 cells.

To further examine the impact of IFITM3 on minimum infectious dosing, we infected A549 lung cells with or without IFITM3 knockdown using H5N1 or H7N3 avian influenza viruses at a range of doses from MOI 0.001 to 10. We observed that infected cells could be detected at lower viral doses for both H5N1 (Fig. 3a) and H7N3 (Fig. 3b) when IFITM3 was deficient. To further examine the effects of IFITM3, we infected both control and IFITM3-knockdown A549 cells with a selection of viruses, including 2 human-origin, 3 swine-origin (2 with known zoonotic transmission^29,30), and 2 avian-origin viruses at an MOI of 1. The cells were incubated with TPCK-trypsin to allow for multiple rounds of infection, and at 48 h post infection, we collected the supernatants to determine viral titers by TCID50 assay. For all viruses, we measured significantly higher titers produced by IFITM3-knockdown versus control cells (Fig. 3c). Along with data from Fig. 1, these data establish that IFITM3 prevents animal-origin influenza virus infection in vitro and in vivo when the virus dose is below a minimum threshold and that influenza virus replicates to higher titers in the absence of IFITM3. These data overall demonstrate that IFITM3 increases the minimum infectious dose necessary to achieve a productive influenza virus infection.

a, b The indicated A549 cells were infected with the indicated avian-origin influenza viruses at a range of 0.001 to 10 multiplicity of infection (MOI) for 24 h. Percent infection was determined by flow cytometry. Data are representative of 3 independent experiments, each performed in triplicate (n = 9). Only comparisons between control and knockdown cells are shown at each dose. c A549 cells were infected with the indicated influenza viruses at an MOI of 1 and incubated in media containing TPCK-Trypsin for 48 h to allow for multi-cycle replication. The supernatants were collected to determine viral titer. Data are representative of 2 independent experiments, each performed in triplicate (n = 6). Only comparisons between control and knockdown cells are shown for each virus. All error bars represent SEM. The numbers shown above the graph represent exact p-values. Comparisons were analyzed by one-way ANOVA followed by Tukey’s multiple comparisons test. Source data are provided as a Source Data file.

IFITM3 deficiency facilitates the adaptation of influenza viruses to a new host species in vivo

Several factors have been proposed to favor inter-species adaptation of a virus to a new host, including levels of virus replication and infectious doses. Given that IFITM3 deficiency impacts both of these features, we tested whether its absence allows influenza virus to adapt more readily in vivo. With biosafety in mind, we did not attempt to adapt avian influenza virus to a mammalian host since such adaptations could increase virus replication across mammalian species, including humans, and could thus be considered dual-use research of concern³². Rather, we sought to adapt viruses of human origin to mice, which has been performed safely for nearly a century in the influenza virus research field³². Initially, we performed lung-to-lung passaging of influenza virus strain A/Victoria/361/2011 (H3N2) 10 times through WT or Ifitm3^−/− mice using intranasal infection and allowing 3 days for the virus to replicate between successive passages (Fig. 4a). We chose this strain because preliminary experiments indicated that it infects mice, but replicates to lower levels than commonly used mouse-adapted strains, thus providing a virus with significant potential for adaptation to mice.

a Schematic of mouse passaging experiments. Initial intranasal infections were performed with 1000 TCID50 of parental viruses. b Schematic of WT mouse challenge with parental A/Victoria/361/2011 (H3N2) or passaged viruses. c–h Groups of WT mice (n = 5 per group) were challenged with 1000 TCID₅₀ of A/Victoria/361/2011 (H3N2) virus passaged 1, 5, or 10 times through WT or Ifitm3^−/− mice and compared to the parent virus (passage 0). c, f Viral titers from lung homogenates collected at day 7 (c represents 2 independent infections). d, g ELISA quantification of IL-6. e, h ELISA quantification of IFNβ. c–h Error bars represent SEM and comparisons were analyzed by one-way ANOVA followed by Tukey’s multiple comparisons test. Each dot represents an individual mouse. The numbers above the graph represent exact p-values. Source data are provided as a Source Data file. a, b Created in BioRender. Denz, P. (2024) BioRender.com/h97v852.

We propagated passages 1, 5, and 10 from both WT and Ifitm3^−/− mice, as well as the parental virus stock (passage 0), in MDCK cells. We then used these expanded stocks to challenge WT mice with equal virus doses. 1000 TCID50 of passages 1, 5, and 10 from WT and KO mice, along with parental virus (passage 0), were used to infect WT mice for 7 days, at which point mice were sacrificed to measure viral titers and inflammatory cytokines (Fig. 4b). To reiterate, passaged viruses, whether generated in WT or KO mice, were tested for adaptation in WT animals. We found that, as expected, the ability to replicate in mouse lungs was unchanged for the earliest passage of virus from WT or KO mice as compared to the parental virus (Fig. 4c). However, passage 5 from KO mice exhibited an upward trend in viral titers and KO passage 10 showed a statistically significant, >1 log average increase in lung viral titers (Fig. 4c). In contrast, virus passages 5 and 10 from WT mice did not show significantly enhanced replication capacity compared to the parental virus (Fig. 4c). As an independent metric for corroborating virus replication, we measured levels of the inflammatory cytokines IL-6 and IFNβ in lungs during infection, which mirrored viral loads, with IFITM3 KO passage 10 inducing a statistically significant increase in both IL-6 (Fig. 4d) and IFNβ (Fig. 4e) compared to infection with parent virus. Since virus adaptation is a stochastic process, we independently repeated our H3N2 virus passaging and testing a second time and obtained similar results indicating enhanced adaptation occurring in IFITM3 KOs (Fig. 4f–h). Interestingly, the animals infected with both of the H3N2 virus passage series did not lose weight despite increased viral titers seen for the IFITM3 KO-passaged virus stocks, suggesting only partial adaptation. We also tested the passaged virus stocks for replication fitness in A549 human lung cells versus LET1 mouse lung cells. We found that the IFITM3 KO-passaged viruses showed increased replication in the murine LET1 cells compared to the parent virus, but replication in human A549 cells was not changed (Supplementary Fig. 7). These results are in agreement with expectations that mouse adaptation would not provide a gain of function for infection and replication in human cells. Overall, our data from both passage series demonstrate that influenza virus gained enhanced replication capacity and induction of inflammation when passed through IFITM3 deficient versus WT hosts.

We next tested whether IFITM3 deficiency is unique in its ability to facilitate virus adaptation or whether passaging in another immune-compromised system that allows enhanced virus replication would also promote interspecies adaptation. The interferon system is particularly important in this regard as interferons widely inhibit virus replication³³. Further, it has become increasingly apparent throughout the recent COVID-19 pandemic that genetic and antibody-mediated interferon deficiencies are significantly more common in the human population than previously appreciated^34,35. We thus repeated our passaging and challenge regimen with influenza virus A/Victoria/361/2011 (H3N2), comparing passaging through WT versus Stat1^−/− mice, which lack signaling downstream of all interferon receptors (Supplementary Fig. 8a, b). Consistent with our previous results, we did not observe significant adaptation of the virus passed through WT mice (Supplementary Fig. 8c). In contrast, passage 5 from the STAT1 deficient mice exhibited a downward trend in viral replication and Stat1^−/− passage 10 showed a statistically significant reduction in the ability to replicate in WT mice (Supplementary Fig. 8c). The titer data were supported by measurement of the inflammatory cytokine IL-6 in that STAT1 KO passage 10 showed a significant decrease in IL-6 levels compared to other groups (Supplementary Fig. 8d). Remarkably, IL-6 levels were uncoupled from IFNβ levels (Supplementary Fig. 8e), which were highest for the STAT1 KO-passaged virus, suggesting that this stock induces increased IFNβ, which may account for decreased virus replication. Additionally, when testing the passaged virus’s ability to replicate in A549 versus LET1 cells, we saw that STAT1 KO-passaged viruses showed reduced replication in LET1 cells compared to the parent virus, with two out of six samples showing replication below detection limits, while replication in A549 cells was unchanged (Supplementary Fig. 8f, g) Our findings are consistent with the principle that, in the absence of selective pressures from the interferon system, viral interferon antagonism mechanisms become less active, resulting in virus attenuation when infecting WT systems. Importantly, these results underscore how IFITM3 may be a unique vulnerability in innate immune response pathways that, when impaired, can promote viral adaptation to new hosts rather than viral attenuation.

To further investigate the generality of our findings on IFITM3, we repeated the passaging regimen in WT and Ifitm3^−/− mice using pandemic influenza virus strain A/California/04/2009 (H1N1), followed by adaptation testing in WT mice. Both WT passage 5 and 10 virus stocks were modestly enhanced in replicative capacity (Fig. 5a). We saw that both WT and KO passage 10 virus stocks induced increased weight loss in infected mice compared to the parent H1N1, with the IFITM3 KO passage 10 mice inducing the most dramatic loss of body weight (Fig. 5b). WT passage 5 and 10 virus stocks exhibited no significant changes in their IL-6 and IFNβ induction capacity (Fig. 5c, d). Consistent with our previous results, virus passaged 10 times in Ifitm3^−/− mice replicated to higher titers compared to the parental strain and to the WT mouse passaged viruses (Fig. 5a). In addition to the highest viral titer, KO passage 10 also induced robust levels of IL-6 and IFNβ in infected lungs (Fig. 5c, d). In a second series of passages with the H1N1 virus, the most robust adaptation in terms of viral replication, weight loss, and inflammatory cytokine induction was again seen for KO passage 10 (Fig. 5e–h). Notably, this adapted viral stock became highly virulent, with all infected mice meeting humane endpoint criteria of >30% weight loss by day 6 post infection (Fig. 5f). Similar to our results with the H3N2 passages, infecting human A549 versus murine LET1 lung cells with the H1N1 passages showed increased replication and IFNβ induction in LET1 cells for the IFITM3 KO-passaged virus whereas we did not observe increased replication in the human cells (Supplementary Fig. 9), again confirming that mouse adaptation of human influenza viruses does not result in a gain of function in terms of increasing virus replication in human cells.

Influenza virus A/California/04/2009 (H1N1) was passaged through mice as described in Fig. 4a. Groups of WT mice (n = 5 per group) were challenged with 1000 TCID₅₀ of A/California/04/2009 (H1N1) virus passaged 1, 5, or 10 times through WT or Ifitm3^−/− mice and compared to the parent virus (passage 0). a, e Viral titers from lung homogenates collected at day 7 (a) or day 6 (e) post infection. Error bars represent SEM, comparisons were analyzed by one-way ANOVA followed by Tukey’s multiple comparisons test. b Weight loss for the H1N1 series 1 challenge. Error bars represent SEM, comparisons were made using the Mann-Whitney test. ELISA quantification of IL-6 (c, g) and IFNβ (d, h) levels in lung homogenates of WT and IFITM3 KO mice at day 7 (c, d) or day 6 (g, h) post infection. Error bars represent SEM and comparisons were analyzed by one-way ANOVA followed by Tukey’s multiple comparisons test. f Weight loss for the H1N1 series 2 challenge. Skull and crossbones indicate humane euthanasia of all animals infected with KO passage 10. Error bars represent SEM, comparisons were made using the Mann-Whitney test. (a, c, d, e, g, h) Each dot represents an individual mouse. b, f dots represent averages of individual mice (n = 5 per group). All numbers above the graphs represent exact p-values. Source data are provided as a Source Data file.

To investigate the specific changes within the passaged viruses that may contribute to the observed mouse adaptation, we extracted viral RNA and sequenced the genomes of passages 5 and 10 from all H1N1 and H3N2 series with comparison to parental viruses. Despite enhanced replication of the H3N2 IFITM3 KO passage 10 viruses, we did not observe changes to the consensus sequence of those viruses ( > 50% prevalence). Instead, we observed the emergence of minor variants, particularly in polymerase segments, that may contribute to the observed phenotypes (Supplementary Data 1, 2). Notably, the second H3N2 passage series showed a trend toward adaptation for the virus passaged 10 times through WT mice in terms of virus replication, though this was not to the same magnitude as the IFITM3 KO-passaged virus and was not statistically significant. This may also be due to the contribution of minor variants present within the quaispecies of that passaged virus (Supplementary Data 1 and 2). The H3N2 virus passaged through Stat1^−/− mice also showed the emergence of minor variants in polymerase segments that are distinct from those present in the IFITM3 KO-passaged viruses (Supplementary Data 1 and 3). These results are consistent with the H3N2 stocks being in the early stages of adaptation as supported by a lack of weight loss when infecting with these viruses and titers that, while increased for IFITM3 KO-passaged viruses, did not reach the levels achieved by the adapted H1N1 stocks.

In contrast to the H3N2 sequences, the H1N1 passages showed clear changes within the consensus sequences in addition to minor variants (Supplementary Data 4 and 5). The trend toward enhanced replication of WT passages 5 and 10 in passage series 1 was associated with two amino acid changes in the viral NP (D101G, R102G) and a single change in PB2 (E158G) that is a known polymerase activity-increasing adaptive mutation in mice^36,37. The H1N1 series 1 KO passages 5 and 10 possessed the same adaptive PB2 E158G mutation seen in WT passages 5 and 10. However, the heightened replication and increased induction of inflammatory cytokines by KO passage 10 relative to KO passage 5 was associated with a single change in the PA polymerase subunit (N373H), representing a novel mutation not previously reported in mouse adaptation studies (Supplementary Data 4). The high virulence of H1N1 series 2 KO passage 10 was linked to a single consensus sequence change, specifically an E158A mutation in PB2 (Supplementary Data 5) that, similarly to the PB2 E158G mutation detected in our first series of H1N1 passages, is a known mouse-adaptive mutation³⁷. Overall, our examinations of replication levels and weight loss, along with sequencing data, suggest that we’ve captured the two viruses (H3N2 and H1N1) at different stages of adaptation, though in each case, this process occurs more rapidly in IFITM3 KO mice. Together with data for the H3N2 virus, our H1N1 passaging experiments demonstrate that the adaptation of influenza viruses to a new host species is accelerated in the absence of IFITM3, fueled by mutations that readily emerge in the compromised host.