Zika virus (ZIKV) is a flavivirus that was initially isolated from rhesus monkeys in 1947 and was first reported in humans in 1952 ¹ The flavivirus genus contains around 70 viruses belonging to the family Flaviviridiae, a family of positive sense, single-stranded, enveloped RNA viruses ² , that also include Dengue virus, West nile virus (WNV), tick-borne encephalitis virus, Japanese encephalitis virus, yellow fever virus, and Hepatitis C virus ³ . Until recently, ZIKV reports had been limited to Africa and Asia ⁴ but it became a wide-spread ZIKV epidemic ⁵ . The virus spread rapidly across the Americas and was declared a ‘global emergency’ by the World Health Organization ⁶ . It is mostly transmitted by mosquitoes and clinical manifestations include rash, mild fever, arthralgia, conjunctivitis, myalgia, and headaches. In addition, the virus can be transmitted sexually, with the risk of infection persisting for several months after initial contact ⁷ . While the symptoms of ZIKV can be mild ¹ , the virus has been linked to two more serious afflictions: Guillen-Barré syndrome (GBS) ^{8, 9} and microcephaly ^{10– 14} , both of which are serious neurological conditions. Microcephaly results in reduced head circumference measurement in infants, exhibiting complications in brain development. Of particular concern is the attribution of microcephaly to infection with ZIKV occurring between the first two trimesters of pregnancy ^{13, 14} . Evidence linking ZIKV to microcephaly includes detection of ZIKV RNA in tissue such as the placenta and amniotic fluid of pregnant women with ZIKV, as well as in the brains of stillborn infants with microcephaly ¹⁵ . In a study with human induced pluripotency stem cells, the mechanism of ZIKV related cell death has been elucidated. This study demonstrated that ZIKV infects human embryonic cortical neural progenitor cells (hNPCs), ultimately leading to attenuated population growth mediated by virally induced caspase-3-mediated apoptosis and cell-cycle dysregulation ¹⁶ . Furthermore, mice studies showed that ZIKV infection can lead to nerve degeneration, softening of the brain and porencephaly ¹⁷ . Additional studies have assessed the causal relation between ZIKV infection and birth defects, and the role of ZIKV as a cause of congenital defects and as a trigger of GBS has been established ^{10, 18, 19} .

Recently, it was discovered that two small membrane-associated interferon inducible transmembrane proteins (IFITMs) IFITM1 and IFITM3 play a protective role against ZIKV infection by inhibiting replication of the virus and preventing cell death ⁷ . The IFITM protein family belongs to a group of small (10 – 15kDa) ²⁰ interferon stimulated genes (ISGs), which are in turn produced by the interferon system ²¹ . The interferon system is the host’s primary response against infection which restricts entry and fusion from late endosomes ²² , with some ISGs being involved in suppressing the early stages of viral replication ^{21, 23, 24} . IFITMs are located at the cell plasma and endosomal membranes, which are the means of entry of many viruses ²⁵ . This protein family protects the host against viral infection by directly restricting entry and fusion from late endosomes ²⁶ .

Flaviviruses depend on the intracellular membrane of their host throughout their lifecycle ^{27– 29} . Mice that are lacking immune sensors, signaling pathways and effector molecules were found to be susceptible to flavivirus infection highlighting the importance of the host’s innate immune response to these viruses ^{5, 30, 31} . Recent literature has demonstrated that during ZIKV infection, the production of several types of interferon and ISGs is increased, and that IFITM1 and IFITM3 are involved in inhibiting ZIKV infection during the early stages of pathogenesis ^{7, 30– 33} . Enveloped viruses, such as ZIKV, have a membrane of host cell lipids that contains viral fusion proteins; these proteins mediate fusion between the virus and the endosomal membrane of the host cell, a necessary step for initiating infection ^{21, 34} . The precise mechanisms by which IFITMs inhibit infection are yet to be described. However, IFITM3 has been shown to impact the function of vesicle-associated membrane protein (VAMP)-associated protein A (VAPA) and oxysterol-binding protein (OSBP) by directly interacting with VAPA. This leads to the accumulation of cholesterol in multivesicular bodies and late endosomes, preventing the fusion of the intraluminal virion-containing vesicles with endosomal membranes ³⁵ . Additionally, IFITM1 has been shown to suppress cell to cell fusion (syncytia formation) by localizing to the plasma membrane after interferon induction ²¹ .

There are additional reasons to explore the role of IFITMs in ZIKV restriction. For instance, murine models of ZIKV infection require deficiency of type 1 interferon signaling, which suggests a role for ISGs in restricting infection ^{7, 36} . Therefore, prior to the induction of ISGs, IFITMs may provide initial defense against the infection ⁷ . Furthermore, IFITMs have been found to restrict virus replication among other flaviviruses, including West Nile virus and Dengue virus ^{23, 37, 38} . Recently, the role of IFITM3 as an antiviral protein against West Nile virus has been explored in vivo, and Ifitm3 ^-/- mice exhibited a greater susceptibility to lethal viral infection, with a greater accumulation of viral protein in peripheral organs and central nervous system tissues ²⁶ . Additionally, IFITM1, IFITM2, and IFITM3 have been shown to have antiviral properties in humans. Polymorphisms in human IFITM3 correlate with the severity of influenza A infection ³⁸ . These functional members are expressed across a variety of tissues in humans and out of the three, IFITM3 is thought to be the most potent line of defense against viral infection ²¹ .

As the mechanism through which IFITM1 and IFITM3 mediate restriction is unknown, computational methods could accelerate research by presenting testable hypotheses. Protein-protein interactions (PPIs) prove to be valuable in understanding the function of a protein, and specifically in how it plays a role in causing or preventing disease. Motivated by this, we had developed a computational model called ‘High-confidence Protein-Protein Interaction Prediction’ (HiPPIP) model that identifies novel PPIs in the human interactome ³⁹ using machine learning to classify features of protein-pairs such as colocalization, coexpression, shared molecular function and biological processes. HiPPIP was also instrumental in discovering that oligoadenylate synthetase like protein (OASL) interacts with retinoic acid inducible gene I product (RIG-I) to activate the RIG-I immunity pathway during influenza viral infection inhibiting virus replication ⁴⁰ . Functional studies initiated solely by this predicted PPI showed that human OASL binds to dsRNA to enhance RIG-I signaling, and that boosting OASL can help inhibit viral infection ⁴⁰ . Using novel PPIs predicted with HiPPIP, we could explain the apparent discordance between modern and historical genetic basis of schizophrenia ⁴¹ , and the role of cilia in the pathogenesis of congenital heart disease ⁴² . PPIs predicted by our method revealed a molecular basis for the negative association between schizophrenia and rheumatoid arthritis ⁴³ . These successes demonstrate that there is enormous potential for biomedical discovery buried in the largely-unexplored novel PPIs in the human interactome.

In this work, we applied the HiPPIP model to discover novel PPIs of IFITM1 and IFITM3, to potentially accelerate the discovery of the mechanism by which they inhibit ZIKV and other viral infections.

We assembled the PPIs of IFITM1 and IFITM3 (‘IFITM interactome’) by predicting novel PPIs with HiPPIP ³⁹ and collecting known PPIs from the Human Protein Reference Database ( HPRD) ⁴⁴ and Biological General Repository for Interaction Datasets ( BioGRID) ⁴⁵ . HiPPIP uses a score cut-off of 0.5 to achieve a high precision of 97.5%, albeit successfully predicting only a few PPIs (recall of 5%), when evaluated on a held-out test data. Thus, the novel PPIs predicted by HiPPIP are highly dependable to lead to successful experiments. Furthermore, predicted PPIs with scores ranging from 0.41 to 0.65 were experimentally validated and found to be true interacting pairs ³⁹ . The HiPPIP model was also computationally evaluated on hub proteins and showed a better performance when compared to Qi et al.’s model ^{39, 46} . HiPPIP is configured to present protein-pairs achieving a score of greater than 0.5 to be novel PPIs. Additional information regarding experimental validation of predicted interactions and model evaluation are available in prior work ³⁹ .

IFITM interactome figure was created using Cytoscape ⁴⁷ . Pathways associated with proteins in the interactome were collected using Ingenuity Pathway Analysis® suite ( www.ingenuity.com). Gene Ontology (GO) terms enriched in the interactome were computed using the BiNGO plugin of Cytoscape ⁴⁸ .

We assembled the PPIs of IFITM1 and IFITM3 (‘IFITM interactome’, Figure 1) by computing novel PPIs using the HiPPIP model and collecting known PPIs from publicly available databases. We found that the interactors of IFITMs are involved in relevant functions including immunity. Functional annotations of the novel interactors of IFITM1 and IFITM3, sourced from Wiki-Pi ⁴⁹ , revealed two main categories: development and innate immunity.

Pathways associated with IFITM interactome computed with Ingenuity Pathway Analysis Suite® are given in Table 1. Gene Ontology biological process terms associated with the interactome, compiled with BiNGO ⁴⁸ are shown in Figure 2 and Table 2.

DEAF1, which encodes for deformed epidermal autoregulatory factor 1 homolog, is a transcription factor that binds to TTCG elements in DNA and is involved in neural tube closure, embryonic skeletal development and anatomic structure morphogenesis, and other functions ⁵⁰ . Additionally, DEAF1 is preferentially expressed in the CNS, particularly during early embryonic development. Mutations in DEAF1 have been associated withintellectual disability ⁵¹ . A study involving Sendai-virus infected mouse embryonic fibroblasts (MEFs) found that DEAF1-/- deficient mice had lower levels of IFN-beta mRNA compared to the wild type controls, along with lower mRNA levels of other markers of viral infection ⁵² .

FNDC3B, a membrane protein, was found to be associated with heart rate, height and corneal structure through genome-wide association studies. While its own functions are unknown, its known interactors are involved in regulation of glial cell apoptotic process, regulation of ion transport (sodium, potassium, calcium) and several cardiac processes.

SPTA1 which encodes for Spectrin alpha, erythrocytic 1 is among the predicted interactors with functional annotations related to innate immunity. It is a molecular scaffold protein that is involved in linking the plasma membrane to the actin cytoskeleton, thereby guiding cell shape and other processes pertaining to the cell’s structure ⁵³ . It is also involved in neural functions of actin filament organization, neurite outgrowth and axon guidance. Considering that IFITM1 and IFITM3 regulate permeability of the cell membrane, it is plausible that SPTA1 is also involved in aiding the host cell’s architecture to impede infection. Another predicted interactor that could be related to the host cell’s defense against infection is RASSF7, which is localized to the microtubule organizing center. While its function is unknown, it has known interactions with proteins that are involved in cell proliferation in the brain, regulation of neuroblast proliferation, nervous system development, synaptic vesicle fusion to presynaptic membrane, and viral budding and assembly.

Other predicted interactors include TSSC4, TLR7, and ARPC1B. TSSC4 is predicted to interact with both IFITM1 and IFITM3. TSSC4’s functions are unknown but its known interactions suggest that it may be involved in viral penetration into host nucleus, protein import into nucleus and immune response signaling, among other processes. TLR7 is involved in several functions and pathways related to innate immunity. It belongs to the Toll-like receptor (TLR) protein family and is involved in the recognition of pathogen-associated molecular patterns (PAMPs), along with stimulating the innate immune response by producing cytokines ⁵⁴ . Its known interactions have cellular components related to membrane components and protein binding. ARPC1B is part of actin related protein 2/3 complex; its known interactions suggest that it may be involved in processes of neuronal development such as axonogensis and development, neuron differentiation, nervous system development, and in immune related processessuch as the innate immune response, regulation of the immune response, etc. These functional annotations are sourced from Wiki-Pi ⁴⁹ ; for example, see DEAF1 at http://severus.dbmi.pitt.edu/schizo-pi/index.php/gene/view/10522.

There is only one study that presents altered gene expression under ZIKV infection available in Gene Expression Omnibus ¹⁶ . The study with eight samples (four infected and four control samples) showed that the infection of human neural progenitor cells (hNPCs) with the virus caused increased cell death and cell-cycle dysregulation ¹⁶ . We examined whether any of the interacting genes were differentially expressed in that study and found five genes that were differentially expressed with a small fold-change but with significant p-value (< 0.005) ( Table 2): CD81, NME5, and RASSF7 were found to be under-expressed and FNDC3B and UIMC1 were found to be over-expressed ( Table 3).

Our goal here is to release the novel PPIs of IFITMs that we could predict computationally, recognizing their importance in mediating immune response in ZIKV infection. These predictions were obtained by using an experimentally and computationally validated algorithm HiPPIP, and are estimated to be highly accurate. Novel PPIs of IFITM1 and IFITM3 shed light on possible mechanisms of host invasion adopted by Zika virus including suppression of the immune system and cognitive and birth defects following infection. Based on the predicted protein-protein interactions, and the functions of the interacting proteins, we formulated three hypotheses about mechanisms of Zika virus infection.

See http://severus.dbmi.pitt.edu/schizo-pi or http://severus.dbmi.pitt.edu/wiki-pi for annotations compiled from various databases for each of the individual proteins.

All pertaining data are provided in the manuscript.