use prefix []or [-]not [+]and [=]has feature [!]exclude feature ie. 'interleukin-6 -animal +phenotypic =protein !tumor'

Displaying 10 papers, 5 pages, start at 1, 417 Hits
68 section matches

Abstract

Oncolytic viruses (OVs) are engineered and/or evolved to propagate selectively in cancerous tissues. They have a dual mechanism of action; direct killing of infected cancer cells crossprimes anticancer immunity to boost the killing of uninfected cancer cells. The goal of the field is to develop OVs that are easily manufactured, efficiently delivered to disseminated sites of cancer growth, undergo rapid intratumoral spread, selectively kill tumor cells, cause no collateral damage and pose no risk of transmission in the population. Here we discuss the many virus engineering strategies that are being pursued to optimize delivery, intratumoral spread and safety of OVs derived from different virus families. With continued progress, OVs have the potential to transform the paradigm of cancer care.

viruses as anticancer drugs

It is well established in several rodent cancer models that a single dose of an effective OV can completely cure disease [8, 9] . This has been shown for DNA and RNA viruses in diverse tumor models. However, while the single shot cure is an exciting prospect for cancer therapy, to date clinical outcomes have typically fallen short of this, and repeat IT virus administration has proven to be a more reliable approach [6, 10] . But there are a number of anecdotal case reports that give credence to the idea that a single shot OV cure may be achievable in clinical practice [4, 11] suggesting that OVs have the potential to transform the practice of oncology.

• • Fusogenic membrane glycoproteins

Fusion of the lipid envelope of an incoming virus with the limiting membrane of its target cell is a necessary step in the life cycle of an enveloped virus, to deliver the encapsidated viral genome into the target cell cytoplasm. This virus-to-cell fusion reaction is mediated by FMGs embedded in the envelope of the virus, and may occur at the cell surface at neutral pH, or in the endo-somal compartment at acidic pH [150] . Neutral pH fusion is triggered by receptor attachment. Cells expressing certain virally encoded, neutral pH-active FMGs on their surface may fuse with neighboring receptor positive cells (cell-to-cell fusion), giving rise to multinucleated syncytia, the hallmark cytopathic signature of a fusogenic virus (e.g., measles) [151] . But more often the FMG is activated only after its incorporation into a budding virus particle, so cannot cause cell-tocell fusion, and must be modified, for example, by cytoplasmic tail truncation, to render it constitutively fusogenic. Either way, FMG-driven fusion of OV-infected cells with uninfected neighboring cells leads to increased bystander killing because multinucleated syncytia are nonviable as well as being highly immunogenic [152] . Based on these observations, nonfusogenic OVs have been armed with FMGs thereby conferring superior oncolytic potency [123, 153] . In one example, an oncolytic herpes virus was rendered highly fusogenic when engineered to encode a cytoplasmically truncated gibbon ape leukemia virus envelope glycoprotein [154] [155] [156] while in another study, an oncolytic VSV was rendered highly fusogenic by replacing its surface glycoprotein (G) with the hemagglutinin and fusion glycoproteins of measles virus.

• Limiting transmissibility

Because the risks of OV transmission are typically unknown when first-in-human Phase I trials are initiated, it is usual to implement standard infection containment measures throughout these studies and to carefully monitor body fluids for the appearance and disappearance of viral genomes and infectious virus progeny [191] . For certain viruses, containment may be considered unnecessary, particularly when there is already widespread population immunity to the OV in question (e.g., measles) or when there has been extensive human experience of exposure to a related virus in the form of a live viral vaccine (e.g., vaccinia). Contingency plans may also be required calling for quarantine of OV-treated patients if treatment toxicity is associated with a longer period of shedding. In reality, to date there has been no instance in which transmission of an OV from a patient to a caregiver or other contact has been demonstrated, and there are no examples of long-term virus persistence or shedding in a treated patient.

engineering viruses: a diversity of platforms

In nature, viruses are continuously evolving and adapting to occupy almost every imaginable biological niche. Viruses infect bacterial, archaea, protist, fungal, plant and animal cells. Their genomes are composed of DNA or RNA, which may be single or double stranded, positive or negative sense, ranging in size from 2 to 300 kb, in complexity from 1 to 300 genes and in capacity for foreign genetic material from a few hundred bases to several kilobases. The particles of viruses from different families also vary enormously in size and structure, ranging from 20 to 1000 nm, from icosahedral to helical symmetry, with or without lipid envelope, integument or matrix and with variable susceptibility to physical disruption. Naturally occurring viruses also offer a vast diversity of virus life cycles, cell entry and replication mechanisms, cell and species tropisms, cycle times, burst sizes, innate immune evasion, apoptosis, antiviral state prevention and immune combat strategies, modes of transmission and pathogenic mechanisms. The diversity of viruses that have been investigated as oncolytic platforms and the different strategies used to improve their efficacy continue to be expanded (Table 1) .

Delivery of Ovs

Mouse models with implanted flank tumors provide a convenient model for IT injection as the tumor is superficial, but in human patients with spontaneous malignancies tumors often develop internally in visceral organs. Melanoma, head and neck cancer, and lower gastrointestinal malignancies may be better suited for IT injections due to anatomical position, although with advanced interventional radiology using computed tomography or MRI image guidance to place the needle many more tumors may be accessible for IT virotherapy. One additional advantage of IT injection over the IV route is that the threshold concentration of input virus required to initiate a spreading infection in the tumor tissue can be more easily achieved.

• Limiting viral neutralization & clearance

For high seroprevalence viruses, one approach to circumvent viral neutralization by preformed antibodies is to engineer or switch the viral coat proteins. Some OVs (e.g., adenovirus) offer a menu of different serotypes, providing a basis for serotype switching between successive doses of the therapy to avoid antibody neutralization, although this does greatly complicate the product development pathway since each serotype is considered to be a distinct pharmaceutical product [82] . For monotypic viruses such as measles, serotype switching is not an option, but antimeasles antibodies can be circumvented by substituting the surface glycoproteins of measles with those of a related but noncross-reactive morbillivirus such as canine distemper virus [83] . Alternatively, the immunodominant epitopes of the measles surface glycoproteins can be modified by mutating key surface residues to eliminate them, or by introducing glycosylation signals so they are shielded by N-linked glycans [84] . It is worth noting that all of these virus engineering strategies have the potential to alter viral receptor usage and hence tumor cell tropism which may limit their utility [83, 85] .

Ov spread

Engineering tumor specificity can also be achieved by detargeting viruses to ablate unwanted tropisms that can cause off target pathology in normal tissues. MircroRNA targeting is the best example of this approach. For example, miRNA targeting was used to control the tropism of an oncolytic coxsackievirus A21 virus which caused rapid tumor regression followed by fatal myositis in murine models of myeloma and melanoma [119] . Insertion of muscle-specific miRNA targets into the viral genome eliminated muscle toxicity but left the antitumor potency of the virus intact. This was shown to be due to miRNA-mediated recognition and rapid destruction of the viral genome in muscle cells. miRNA targeting has since been applied to many OVs from diverse virus families and provides a convenient and economical strategy (using sequence insertions of only ∼100 bases) to control unwanted virus tropisms [25] .

• viral amplification & spread

In the case of free virus particles, which are susceptible to antibody neutralization, the stroma of the tumor also has the potential to limit IT diffusion and block systemic release. Fibroblasts, endothelial cells, immune cells and the ECM make up the tumor stroma, and the exact composition varies widely depending on tumor type [125] . The ECM is a collagenous matrix of protein fibrils, adhesive proteins and proteoglycans that create a web with pore sizes similar or slightly smaller than virions [126] . Destruction of ECM components can facilitate viral spread throughout the tumor and can be achieved using conventional chemoradiotherapy or by delivering matrix-degrading enzymes such as collagenase and hyaluronidase [127] . An alternative approach is to encode a matrix-degrading enzyme or inducer of matrixdegrading enzymes in the viral genome. For example, hyaluronidase and relaxin encoding oncolytic adenoviruses have each been shown to spread more efficiently in experimental tumors [128] [129] [130] .
Another factor that significantly impacts the kinetics of virus spread is the burst size, or the number of progeny viruses released by a productively infected cell, which varies widely between viral families. Picornaviruses, VSV and vaccinia virus can release up to 10,000 progeny from a single infected cell after a delay of only 6-18 h [131] [132] [133] . In addition to innate antiviral immunity, adaptive cell-mediated immune responses are typically required for the complete elimination of a viral infection and act by eliminating infected cells before progeny can be released. Oncolytic virotherapy can therefore be viewed as a race between the spreading virus and the responding immune system. For this reason, faster moving viral infections are often considered capable of inflicting greater damage to an infected tumor before they can be contained by the immune system [134] . However, in defense of the viruses with smaller burst sizes, or which release progeny by budding, they tend to be less review Maroun, Muñoz-Alía, Ammayappan, Schulze, Peng & Russell future science group rapidly controlled both by the innate and adaptive host immune responses [135] . Therefore, as with the classic race between the hare and the tortoise, it is very difficult to predict whether a fast or slow replicating virus will show superior efficacy in a given preclinical cancer model.

Arming Ovs: extending the range

No matter how extensively an OV infection spreads through a tumor, a sizable percentage of the cancer cells will escape infection [61] . Killing of these uninfected (bystander) cancer cells is therefore critical if oncolytic virotherapy is to become a curative, as opposed to just a tumor debulking strategy. Bystander killing can be achieved, both locally at the site of a spreading infection and systemically at uninfected tumor sites, by genetically arming the virus using one of several possible approaches. For example, an OV can be engineered to encode a secreted protein that selectively mediates the destruction of neighboring and/or distant cancer cells. Alternatively, it can be armed with a 'suicide gene' or prodrug convertase whose encoded protein converts a harmless prodrug to a diffusible anticancer drug. In a third approach, the OV can be armed with a 'radioconcentrator gene' so that infected tumor cells are able to concentrate a β-emitting radioisotope whose emitted electrons damage adjacent uninfected tumor cells. Fourth, it can be engineered to fuse infected tumor cells with uninfected neighboring cells. And last, but by no means least, it can be engineered to express one or more genes capable of amplifying immune-mediated killing of uninfected tumor cells. Each of these approaches is discussed below using specific examples to illustrate the concepts.

• • Limiting pathogenicity

Understanding the selective pressures that operate within the tumor and the host, as well as the role of viral quasispecies in treatment outcomes is an active area of research. RNA and DNA viruses exist as a population of quasispecies or collections of related viral genomes undergoing variation and selective pressure [174] . Generation of quasispecies occurs when viral genomes are copied during the replication cycle. Viral polymerases typically have an error rate that introduces an average of one or more base mutations per progeny genome [175] . In general, the larger the virus genome, the lower the polymerase error rate. Picornaviruses are among the smallest viruses being developed for oncolytic therapy with positive sense RNA genomes ranging from 7 to 9 kb in length, and their RNA polymerases have a correspondingly high intrinsic error rate. Production of a virus for clinical application is a highly regulated process and involves multiple rounds of replication to achieve enough virus for patients. The viral product is therefore already a swarm of quasispecies at the time it is administered to the patient and mutates further as it undergoes additional rounds of replication in vivo. Studies of the mutation rates of viral polymerases, the generation of quasi species, the evolution of viral populations and the evolution between dominant subspecies within a virus population are therefore of great interest and relevance to the OV field.
Polymerase infidelity is not the sole driver of virus genome diversification and evolution. For many viruses, recombination between homologous viral genomes can occur when two related viruses infect the same cell [178] . This type of recombination is not a feature of negative strand RNA virus evolution because the nascent progeny viral genomes are cotranscriptionally incorporated into a helical ribonucleocapsid structure which prevents recombination. However, for positive sense RNA picornaviruses, and DNA viruses such as adenovirus, HSV and vaccinia, it is important to consider the possibility that the input virus may encounter and recombine with a homologous 'wild-type' virus in a treated patient. While such recombination has never been documented in OV trials, there are well described examples in the field of vaccinology, such as the re-emergence of pathogenic polioviruses through the recombination of vaccine genomes and naturally circulating picornavirus genomes [179] . In view of these risks, it is generally considered inadvisable to arm viruses with therapeutic transgenes that have theoretical potential to increase pathogenicity if transferred to a related naturally circulating virus. Genes encoding immunosuppressive, antiapoptotic or directly cytotoxic proteins are on this list.

• Limiting transmissibility

In light of the inconvenience and undesirability of OV shedding, there is interest in engineering strategies that may selectively interfere with the process. As one example, wild-type measles virus is a highly transmissible airborne virus that uses the nectin-4 receptor to enter into airway epithelial cells from whence its progeny are shed into respiratory secretions [192, 193] . Eliminating the nectin-4 tropism by strategically mutating key surface residues in the hemagglutinin attachment protein results in a virus that still causes measles in nonhuman primates, but which is no longer shed into respiratory secretions or urine [194] . Measles virus RNA (but not infectious virus) was detected in the blood, urine and saliva of myeloma patients up to 3 weeks after IV administration of an oncolytic measles virus, especially at higher dose levels, and it is possible that this shedding might be eliminated by using a nectin-4 blind version of the virus [195] . Virus shedding has also been detected in human clinical trials of HSV, reovirus, vaccinia, reovirus and adenovirus oncolytics, but was less readily detected following IT therapy with oncolytic VSV and polioviruses [167] . However, future science group www.futuremedicine.com it should be noted that while genome sequences have been detected in shed material, infectious virus particles have not been recovered.
An alternative strategy often used to side step antibody neutralization, at least of the first dose of virus administered, is to use OVs derived from zoonotic animal viruses such as New Castle disease virus (chicken), VSV (cattle), myxomavirus (rabbit), Seneca Valley virus (pig) or mengovirus (e.g., mouse and monkey). Additional regulatory executive summary viruses as anticancer drugs • Viruses naturally possess many properties that favor infection of cancer cells. Enhancing these natural properties and adding new properties through directed evolution and genetic engineering are used to create oncolytic viruses (OVs), which are emerging as a new anticancer drug class.

viral spread

• Targeting viral spread to tumor cells can be accomplished by transductional targeting (modifying receptor tropism), transcriptional targeting (controlling virus gene expression with tumor-specific promoters), physiologic targeting (disrupting viral immune combat proteins), apoptosis targeting (disrupting viral antiapoptotic proteins) or miRNA targeting.

viruses as anticancer drugs

The key desirable characteristics of any OV are specificity, potency and safety; specificity for the targeted cancer, potency to kill infected cells and cross-prime antitumor immunity, and safety to avoid adverse reactions and pathogenic reversion.

Delivery of Ovs

Viral infections are well understood but for OV applications must be viewed as drugs obeying pharmacological principles. A drug is typically administered to a patient in an extremely controlled way to produce a reliable, consistent and predictable pharmacokinetic profile (absorption, biodistribution, metabolism and excretion) and bioavailability. Natural viral infections do not obey these rules since the inoculum is of variable size; host resistance varies from person to person; and the kinetics of the adaptive immune response differ greatly between individuals. Thus, the outcomes of natural infections with a given pathogenic virus range from asymptomatic seroconversion to full blown disease. The primary virus inoculum must contain sufficient virus particles to overcome initial host defenses at the site of entry. Some viruses enter through mucous membranes in the GI or respiratory tracts, while others enter via direct inoculation into the blood stream following a needle stick or arthropod blood meal [58] . OVs are delivered in the same way as traditional drugs, by introducing a highly concentrated virus inoculum into the body via oral, intravenous (IV), intranasal, transdermal, subcutaneous or intramuscular routes whereupon the dispersion of the inoculated virus, or its progeny, takes it to the targeted cancerous tissues. A major factor distinguishing OVs from traditional drugs is that they self-amplify and spread after delivery so their peak concentration may not be reached until sometime after the treatment is administered. Biological amplification by viral replication is the most important difference between viral therapies and traditional drugs. The concentration of a drug diminishes over time at a very well described rate depending on the clearance and elimination from the body. Through the course of a viral infection the viral load is initially small and then increases and finally decreases rather than just following a specific rate law of elimination. Limited amounts of virus particles replicate at the site of inoculation [59] . Input or progeny viruses then can either drain with the lymphatic fluid to the nearest lymph node, go directly into circulation or spread locally before spreading systemically to eventually arrive at a specific target organ. Pathology is induced by high levels of viral replication in the target organ directly killing infected cells or recruiting the immune system to kill them, at the same time provoking a local inflammatory response.

• • Limiting pathogenicity

Although it will be interesting to determine whether OVs encoding modified polymerases with higher fidelity will exhibit an improved safety profile [176] , it should be noted that OV reversion to a more pathogenic state has not yet been observed clinically. This is also true of many time-honored viral vaccines. For example, in 50 years of widespread measles vaccination using attenuated Edmonston lineage substrains, there has never been a documented reversion back to a wild-type phenotype, even after IV doses as high as 10 11 TCID 50 were administered to immunosuppressed, measles antibody negative myeloma patients in an oncolytic virotherapy study [177] .
Certain viruses with segmented genomes, most notably the orthomyxoviruses (e.g., influenza), are capable of rapid evolution by genome fragment reassortment in multiple infected cells. Reassortment of genome fragments in virus-infected birds and pigs is considered to be a key driver of the antigenic shifts that occur during evolution of new pathogenic influenza virus strains [180] . This important safety concern may explain why oncolytic influenza viruses have not yet been advanced to human clinical trials [181] . Because of the toxicity risks associated with the proliferative and evolutionary capacity of OVs, there is considerable interest in contingency plans to terminate the spread and/or transmission of an OV infection. Reliable antiviral medications are available for TK-expressing herpes viruses, but this is not the case for the majority of viruses being developed as OV platforms [182, 183] . Attention has therefore been directed to the development of safety switches that can be engineered into the OV genome and triggered on demand to terminate in vivo replication.
Limiting the pathogenic potential of OVs can also be accomplished by enhancing the host innate immune response. VSV has been engineered to express IFN-β, which induces an antiviral state and reduces proliferation of cells. Cells capable of responding to interferon signaling will limit viral replication in response to the virally produced interferon, further restricting viral replication to tumors with defective innate responses [188] . The expression of IFN-β from infected tumor cells can also prevent off-target infection of nearby tissue through paracrine signaling. Another way to achieve a similar effect is mutating the VSV matrix protein, which is responsible for limiting cellular production of type I interferons [189] . This, and similar approaches, to inactivate viral genes that combat innate immunity has been extensively utilized in the OV field, across a broad range of viral families. Sensitization of OVs to innate immune responses can limit off-target infections and enhance tumor-specific tropism while increasing the therapeutic index.

• • Sodium iodide symporter (radioconcentrator)

Iodide is a critical component of thyroxine and is concentrated in thyroid follicular cells by the thyroidal sodium iodine symporter (NIS), a cell surface glycoprotein with 13 transmembrane domains [146] . Radioiodine is therefore used routinely in the clinic for thyroid imaging and for ablation of overactive thyroid tissue, including metastatic thyroid cancer. This is facilitated by the ready availability of several iodine radioisotopes, most notably 123 Ifor γ camera and single-photon emission computed tomography imaging, 124- for PET imaging and 131 I -, a β-emitting isotope, for thyroid ablation. Besides radioiodine, NIS can concentrate several related anions, of equal or greater value for single-photon emission computed tomography ( 99m TcO 4 -, pertechnetate), PET (B 18 F 4 -, tetrafluoroborate) and tissue ablation ( 211 At -, astatide, 188 ReO 4 -, perrhenate) [147] . The β emissions of 131 Ihave an average path length of approximately 1.8 mm in tissue and can therefore inflict significant damage on cells adjacent to an 131 I --loaded NIS expressing cell. For 188 ReO 4 the β emission path length is longer such that NIS negative tumor cells are even more likely to be damaged in the β particle crossfire when this radioisotope is used. Unsurprisingly, therefore, the NIS gene has been engineered into several OVs whose IT spread has been elegantly mapped and monitored in tumor-bearing mice by serial radiotracer imaging, and whose potency has been substantially boosted by appropriately timed administration of 131 I - [148] . At least two NIS-expressing OVs, a prostate-targeted oncolytic adenovirus and a CD46-targeted measles virus, have been advanced to human clinical trials with positive imaging data reported [11, 149] .

viruses as anticancer drugs

Oncolytic viruses (OVs), evolved and engineered for cancer specificity, are gaining momentum as a new drug class in the fight against cancer. Besides, causing the death of virus-infected cancer cells, the spreading intratumoral (IT) infection can also boost the anticancer immune response, leading to immune destruction of uninfected cancer cells. The paradigm of OVs has been reviewed extensively [1] [2] [3] [4] [5] [6] [7] .
Based on these clinical advances, development of OVs is rapidly accelerating. The purpose of this review is to discuss the virus engineering approaches and OV performance optimization strategies that are being pursued in the field, and to point out some of the challenges that remain. We believe this perspective will be particularly valuable to virologists entering the field. We apologize to the numerous investigators whose work has not been acknowledged; we are well aware of these omissions but due to space constraints our approach has been choosing illustrative examples that demonstrate the key principles. The review is divided into five main sections: (i) an introduction to viruses and virus engineering, (ii) delivery, (iii) spread, (iv) arming and (v) safety. All five topics are highly interdependent.

engineering viruses: a diversity of platforms

Aside from their differences, viruses share fundamental similarities including their dependence on the host cell to provide a suitable environment for genome amplification, gene expression and progeny virus production and their adaptation to a specific set of host cell conditions and factors such that propagation is precluded in cells that fail to provide the necessary environment. These characteristics allow viruses to be targeted to cancer cells as a self-replicating antineoplastic therapy.

Delivery of Ovs

OV therapy begins with the administration of the virus to the patient just as with any other drug. Several approaches can be taken, each with to optimize oncolytic virotherapy indicated that maximizing delivery, so more tumor cells are infected, has a profound impact on the degree to which the virus must propagate from each infected cell for curative therapy [60] . The higher the number and the more random the dispersion of infectious centers established in a tumor after delivery of virus, the more effective the virus will be, although these parameters must be optimized for each virus and model [61] . IV and IT administrations of OVs are common delivery methods that introduce large amounts of virus into one compartment in a very short time. This rapid delivery of a massive virus inoculum does not parallel what typically occurs in a natural infection and potentially allows the IT spread of the OV infection to outpace host defenses. Successful delivery in the context of an OV infection requires that a critical viral concentration is achieved within the tumor to allow for sufficient oncolysis and systemic spread of the virus to all sites of disease in a safe and reproducible pattern.
• • iv delivery IV delivery of an OV seems advantageous in the setting of metastatic disease, allowing the virus to potentially access all sites of disease via the circulation soon after infusion. In a clinical setting, systemic delivery may be preferred since it is more broadly applicable than IT inoculation, regardless of tumor location and total tumor burden. Yet, attempts at systemic delivery have shown limited success; the administered virus is immediately diluted in the circulating blood volume (∼5 l), so extremely high doses are needed to achieve meaningful circulating titers. A clinical trial using an oncolytic pox virus, JX-594, demonstrated that virus could only be detected in tumors of patients that received at least 10 9 p.f.u. per IV dose without any virus being detected in tumors of the lower dose cohorts of patients [62] . Because of this important dilution effect, doses required for IV therapy may have to be up to 1000-fold higher than for IT administration, which results in significant manufacturing challenges as well as a unique set of toxicities [2] . Administration of high titers of virus intravenously can lead to hepatotoxicity, thrombocytopenia and lymphopenia [33, 63] .
Following the path of an OV from the IV line to the tumor will quickly illustrate some of the challenges for IV delivery. First, the virus is rapidly diluted in the circulation, where its infectivity is neutralized by serum proteins including antibodies and complement, and a large proportion is sequestered by reticuloendothelial phagocytes in the spleen and liver. Any virus particles that reach the tumor vasculature without first being neutralized must then extravasate through fenestrations, and pores between the vascular endothelial cells of the tumor capillaries, and once in the interstitial space of the tumor, must negotiate the extracellular matrix (ECM) to reach and infect the resident cancer cell [64] .
Tumor vasculature can both aid and limit viral access. Tumors typically have an unorganized growth pattern, producing vasculature that is tumultuous and insufficient resulting in heterogeneous perfusion throughout the parenchyma [65] . Decreased perfusion limits the formation of infectious centers evenly throughout a tumor. Manipulating physiological parameters such as blood pressure and systemic vascular resistance can preferentially increase tumor perfusion and promote better delivery of viral vectors [66] . Once in a tumor vessel, the virus must move from the intravascular compartment into the interstitium by crossing the vascular endothelium to continue its journey through the interstitial space until finally reaching a permissive tumor cell.
At least in theory, viruses can cross the endothelial lining of tumor neovessels by diffusion, active infection, or by trafficking inside of or on the surface of marrow-derived cells that are capable of diapedesis. Tumors generally have poor lymphatic drainage and can have leaky blood vessels with a wide range of pore sizes which helps to explain the phenomenon of enhanced permeability and retention of nanoparticles such as OVs in the tumor parenchyma [67] . In some tumors, the pores and fenestrations in tumor capillaries may be large enough to permit passive diffusion of viruses (depending on their size) but in many tumors review Maroun, Muñoz-Alía, Ammayappan, Schulze, Peng & Russell future science group the endothelium is sufficiently organized that this type of extravasation is simply not possible.
Another approach to help OVs' extravasate is by temporally creating pores throughout the tumor microvasculature to enhance transport of the virus into tumor interstitium. This can be done through application of focused ultrasound in the presence of a contrast agent or a polymer. The microbubbles or polymers are injected intravenously and can freely circulate without harming tissues, but when they are exposed to ultrasound waves they can oscillate and induce pore formation or cavitation in the surrounding tissue. The ultrasound can be focused on a tumor just after the infusion of the virus to increase viral extravasation. This has been shown to enhance vaccinia virus delivery to tumor xenografts and can increase viral transgene expression by more than a 1000-fold in the tumor [71, 72] . Many strategies are being developed to enhance OV IV delivery.
• • iT delivery IT administration directly delivers a high concentration of the OV into the parenchyma of the injected tumor but may not result in spread of the virus to distant sites of metastasis. T-Vec is administered by IT injection to accessible skin tumors of patients with metastatic malignant melanoma. The injection is repeated every 2 weeks until lesions have resolved or failed to respond [12, 73] . Interestingly, viremia was not documented in patients, suggesting that the regression of noninjected lesions was likely immune-mediated and not due to direct oncolysis [74] .
Immediately following IT delivery, small levels of virus can be detected in the blood from leakage through injured tumor vasculature [75] , but that small amount of virus is quickly cleared from the system. Also, virus that has been injected into a tumor is frequently extruded immediately from the injection track once the syringe is removed, especially in smaller higher pressure tumors. However, systemic viral spread can occur following IT administration if the initially exposed tumor cells are capable of amplifying the input virus and releasing progeny into the bloodstream. Such secondary viremia may not peak until several days after IT virus delivery and, with currently used OVs, is often not detected. Secondary viremia, when it occurs, more closely mimics the natural propagation of a viral infection from the site of inoculation to more distant target organs.

• Other routes of delivery

Intraperitoneal virus administration is often pursued in virotherapy studies aiming to impact ovarian cancer and other disseminated intraperitoneal malignancies, while intrapleural administration is pursued for mesothelioma therapy. Both of these approaches are similar to IT administration in that the virus comes into direct contact with tumor cells in the injected cavity, although there is a greater risk that input virus will be immediately neutralized by antibodies and other proteins in ascites or pleural fluid. Intravesical instillation of virus is the favored route of administration for treatment of early-stage bladder cancer, requiring only that future science group www.futuremedicine.com the input virus be stable in urine. Also, immediately after brain cancer surgery, in an attempt to control residual disease, virus is often instilled directly into the resection cavity.

• Limiting viral neutralization & clearance

Perhaps the most significant barrier to widespread OV delivery is rapid neutralization and clearance of circulating virus particles. Antibodies and complement proteins can coat the virus blocking its ability to interact with its cellular receptor and accelerate Fc receptor-mediated clearance by splenic macrophages and hepatic Kupffer cells.
Mengovirus, VSV and Newcastle disease virus are examples of animal pathogens being developed as oncolytic agents. Their appeal as oncolytic platforms is due to the coupling of their ability to selectively propagate in human tumors with their low seroprevalence in the human population [77] . Unlike viruses such as measles which have almost exclusively human cell tropism, these zoonotic viruses are capable of infecting both rodent and human tumor cells allowing preclinical testing in more informative immunocompetent mouse cancer models.
Virus-infected cell carriers can also be used to transport viruses via the bloodstream to sites of tumor growth. In part, this is possible because there is an 'eclipse period' after the cell carriers have been infected during which they do not display viral proteins on their surface and are therefore not bound by virus neutralizing antibody, but are still able to extravasate from tumor neovessels and release infectious progeny into the tumor parenchyma. Several cell carrier/OV combinations have been used including mesenchymal stem cells, dendritic cells, T cells and endothelial progenitor cells [89] [90] [91] [92] . Some carrier cells are believed to home more efficiently to tumors responding to chemotactic signals arising from hypoxia or IT inflammation.

Ov spread

Replication and amplification in the tumor is the major feature that sets OVs apart from other anticancer drugs. No traditional chemotherapy, immunotherapy or small molecule inhibitor can target tumor cells and then amplify at the site of action and spread to other sites of tumor growth. Specificity as well as speed and extent of IT virus replication are the key determinants of therapeutic index (efficacy/toxicity ratio) for an OV therapeutic, and each of these parameters can be modified by virus engineering. Additionally, the kinetics of spread can be impacted by combining the virus with immunomodulatory drugs. The idea of a virus with selective tropism for cancerous tissue has obvious appeal. Even for naturally occurring viruses that have not been tropism modified, a tumor offers a favorable environment to support a productive infection, and this is borne out by case reports of temporary remission or regression of different cancers concurrent with viral infections [4, 93] . There are several reasons why tumors are generally more susceptible than normal tissues to virus attack; poorly developed lymphatics, high nonsuppressible metabolic activity, resistance to apoptosis, poor responsiveness to interferon and intrinsic suppression of immune effector cells. But as we move in the direction of intentionally using virus infections to mediate tumor destruction, it is apparent that a targeted virus with exquisite tumor specificity will be superior to its non targeted counterpart, allowing for the administration of higher tumor destructive doses without toxicity to normal tissues.
Virus tropism is determined by many factors, most of which, if understood sufficiently, can be manipulated to enhance tumor specificity. The receptor tropisms of naturally occurring viruses are rarely of interest for tumor targeting, but this is not true for tissue-culture-adapted vaccine lineage viruses, some of which have evolved in the laboratory to use receptors more abundantly expressed on cancer cells. Examples include the CD46 receptor tropism of vaccine lineage measles virus and the heparan sulfate tropism of laboratory-adapted Sindbis virus [94, 95] . Where greater specificity is desired, it may be possible to engineer new receptor tropisms by modifying the structure of a viral attachment protein, for example, by displaying polypeptide ligands at the extreme C-terminus of the measles H glycoprotein [96, 97] . However, for many viruses, displaying a polypeptide ligand on the surface does no more than redirect attachment and does not confer a new receptor tropism [98] [99] [100] . This remains an area of active investigation.
In addition to their dependence on specific entry receptors, the surface glycoproteins of many enveloped viruses (and sometimes other critical viral proteins) must be proteolytically activated before they are competent to mediate virus entry [101] . By engineering the protease target sequences in these viral glycoproteins it is possible to generate viruses whose propagation is now dependent on exposure to a specific protease with high IT abundance, such as urokinase, matrix metalloproteinase 1 or cathepsin D [102] [103] [104] .
Beyond the step of cell entry, viruses are exquisite sensors of intracellular processes and can therefore be adapted or engineered in several ways for intracellular targeting of cancer cells. This is best understood by considering some fundamental aspects of the interplay between a virus and an infected cell. The incoming virus aims to usurp the cellular synthetic machinery for generation of progeny viruses. The cell resists this takeover bid by rapidly detecting virus invasion, then triggering a signaling cascade that leads to establishment of an antiviral state and release of interferon which induces an anti viral state in adjacent cells [105, 106] . The antiviral state is very complex but suppression of protein translation is a key component. Apoptosis is also triggered by the virus detection machinery so that the infected cell dies before it is able to manufacture virus progeny [107] .
For a virus to be 'successful' it must combat these host cell responses, avoiding detection, as long as possible, for suppressing the establishment of an antiviral state and preventing apoptosis. Virtually all naturally occurring viruses therefore encode proteins that inhibit apop tosis and the antiviral state [108] . Removing these accessory functions from the viral genome leads to virus attenuation in normal tissues, but to a much lesser degree in cancerous tissues. This is because cancer cells are intrinsically resistant to apoptosis and to the establishment of an antiviral state, making them highly susceptible to attenuated viruses that are no longer able to control those processes. This has provided a mechanistic basis for physiologic targeting of several viruses; VSV by mutating the matrix gene whose encoded protein blocks the interferon response [109] ; HSV by mutating both copies of the γ-34.5 gene which interferes with interferonmediated shutoff of host protein synthesis and enhances neurovirulence [110, 111] ; adenovirus by mutating the E1B protein, one of whose actions is to inhibit the apoptotic activity of p53 [112] .
Viruses can be further engineered to exclusively replicate in tumor cells by combining a virus' needs with physiologic peculiarities intrinsic in tumorigenesis. For example, a virally encoded thymidine kinase (TK) is required for HSV and vaccinia virus infection to increase the supply of deoxynucleotide triphosphates required for synthesis of progeny virus genomes [113, 114] . Elimination of the TK future science group www.futuremedicine.com gene from the viral genome restricts viral replication to cancer cells where there is an upregulation of human TK [115, 116] . Another example of a virus that exploits the high replication rate of tumor cells is Toca-511, a replication competent C-type retrovirus encoding the drug activating enzyme cytosine deaminase (CD) [117] . Since the integration of C-type retroviruses is S-phase dependent [118] , this virus is selectively amplified in rapidly proliferating tumor tissue.

• viral amplification & spread

After an appropriately targeted virus has infected a tumor cell, it is the extent of its subsequent propagation that becomes the key driver of potency. Data-driven mathematical models of systemic oncolytic virotherapy indicate that tumor eradication is dependent on two major parameters: the initial density and distribution of infectious foci in the tumor; and the ultimate size of the infectious centers arising from each individual infected cell (i.e., virus spread) [120] . Viral spread may occur by various mechanisms. Local spread may occur by intercellular fusion, by direct transfer of virus from infected to adjacent cells, or by release and local migration of progeny virions through the interstitial space. Systemic spread as free virus particles or as virusinfected migratory cells occurs via lymphatic channels or via the bloodstream.
Direct cell-to-cell transfer of viruses has the advantage of stealth as the virus cannot be neutralized by antiviral antibodies in the interstitial fluid [121] . Nonfusogenic viruses can be armed with fusogenic membrane glycoproteins (FMGs) to enable stealthy spread through intercellular fusion leading to the formation of large, nonviable multinucleated syncytia which may also serve as excellent antigen presenting cells for amplification of the antitumor immune response [122] . By way of example, VSV encoding measles fusogenic glycoproteins and HSVs encoding the fusogenic gibbon ape leukemia virus glycoproteins have shown superior efficacy when compared with their corresponding parental viruses [123, 124] .

• • Secreted toxins

As a general rule, if a virus is to be armed with a gene encoding a secreted toxin, that toxin should be targeted so that it can kill only cancer cells. In the absence of such targeting, there would be little prospect of avoiding off-target toxicities. Immunotoxins are bifunctional proteins in which plant or bacterial toxins (typically ribosomal inhibitors) are fused to a single chain antibody or other polypeptide domain to target endocytosis in cancer cells [136] . In theory, such molecules that have been extensively investigated and advanced to human clinical trials particularly for the treatment of B-cell and T-cell malignancies [137] , could be expressed from an engineered OV genome.

• • Arming Ovs to amplify antitumor immunity

Each step of the above process can be impacted by virus engineering. Not only can the kinetics and mode of cell killing be manipulated (see the previous section on virus spread), but also the OVs can be armed with a wide variety of immunoregulatory genes whose products will be secreted into the interstitial fluid space. Thus, OVs have been engineered to express the following: high levels of type I interferons to better drive the early innate/inflammatory response; GM-CSF to stimulate the phagocytic activity and lymph node trafficking of professional antigen presenting cells; chemokines to enhance the IT recruitment of immune and inflammatory effector cells, especially cytotoxic T cells; cytokines to drive the activation and proliferation of tumor-resident cytotoxic T cells; bispecific T-cell engagers to enable tumor cell killing by T cells not recognizing tumor antigens; checkpoint inhibitor antibodies to block the protective PD-L1/B7.1 shield that protects tumor cells from T-cell attack; cytokines to increase MHC-peptide neoantigen expression on uninfected tumor cells; or cloned tumor antigens to further amplify tumor-specific immunity [165] .
While each of these approaches has proven beneficial in selected preclinical animal models where immune destruction of the tumor is the dominant OV effect, they are equally likely to decrease potency by speeding virus elimination in situations where the spread of the virus is dominant.

Safety

Engineering viruses for increased stealth, greater specificity, faster spread and enhanced potency is not without risk. As OVs gain new properties, checks and balances must be applied to assure that they do not readily evolve into pathogens, let alone transmissible pathogens that could pose a risk not only to immediate contacts of the treated patient, but also to the greater population.

• • Limiting pathogenicity

Viral cytotoxicity is the basis of tumor cell death necessary for oncolytic activity. Off-target infection and killing of normal cells by poorly targeted OVs or by OVs that have evolved new tropisms in a treated patient can cause unwanted normal tissue pathology. OVs are therefore delicately balanced between retaining enough virulence to substantially decrease tumor burden versus being sufficiently targeted (or attenuated) to not cause a new disease in the patient. Cancer-specific targeting is the most critical safety feature, but viruses evolve and viral populations are dynamic [172] . Evolution is a constant accompaniment of a spreading virus infection, whether or not the virus is oncolytic; in vivo progeny of the therapeutic OV differs on average by a single -point mutation per genome from the input virus [173] . Hence, as the input virus is amplified, it generates a swarm of quasispecies viruses, each one a slightly imperfect replica of the input virus. This swarm of progeny viruses is subjected to selective pressure as it encounters new biological niches in the treated cancer patient. Thus, if it so happens that a member of the swarm is capable of infecting a normal host tissue, the virus may have gained a new foothold from which to further evolve. Gaining new cell tropisms or losing restriction factors is therefore a significant theoretical concern in oncolytic infections, but has not yet been documented in human trials, nor in preclinical models. However, given the importance of this particular scenario, a great deal of attention is paid to the problem not just by investigators, but also by regulatory agencies.

• Limiting transmissibility

For obvious reasons OV transmission from a treated patient to a caregiver, family member, coworker, pet or other species is highly undesir-able. Infectious virus particles may be present in the blood of a treated patient, and may be shed into the environment in urine, feces, saliva and other bodily secretions. Contact with the body fluids of an OV-treated patient therefore may have the potential to spread an infection to new hosts [190] .
In an OV-treated patient pre-existing antiviral immunity can be a formidable barrier to efficacy, particularly if the virus is administered intravenously. However, pre-existing antiviral immunity in caregivers and patient contacts provides reassuring protection against virus transmission. Conversely, when this particular efficacy barrier is circumvented by using OVs engineered for antibody evasion or selected for low seroprevalence, the risk of epidemic spread in the human population, unchecked by pre-existing herd immunity, looms larger.
• The diversity of virus families and engineering techniques allows for the creation of OVs with a wide range of properties that can be tailored for each type of cancer.

Delivery of Ovs

• Intravenous delivery allows a virus to reach distant sites of metastasis via the circulation, but extravasation into the tumor parenchyma is inefficient.
• Intratumoral injection can concentrate virus at a site of tumor growth, but regression of distant tumors requires that the virus spread systemically or induce a systemic antitumor immune response.
• Neutralizing antibodies, hepatosplenic sequestration of the virus by macrophages and dilution of the virus in blood or tissue may limit the effectiveness of treatment.

Safety

• Ideally, OVs should be nontransmissible. scrutiny is generally required for these viruses (e.g., from the US Department of Agriculture) to address the additional risks of environmental release and epidemic spread in domestic animals, particularly as they relate to agricultural livestock.

Conclusion & future perspective

Viruses are at last being harnessed for the benefit of cancer patients. The OV field has moved well beyond proof of principle in human studies, and virus engineering will be the key to its continued advancement in the coming years. Virtually every component of every naturally occurring or laboratory-adapted virus can be engineered and/or evolved to enhance its suitability for cancer therapy and we are currently witnessing unstoppable creative activity in this area. Safety is obviously of paramount importance in this relatively new field, and is therefore closely regulated from the design stage to clinical implementation. Considering the current trajectory of OV research, there can be little doubt that viruses are on their way to becoming one of the foundational modalities of future cancer treatment regimens.

viruses as anticancer drugs

In light of recent clinical progress, interest in the approach is burgeoning. One critical milestone was the 2015 marketing approval granted in Europe and the USA for talimogene laherparepvec (T-Vec, Imlygic™), an engineered HSV encoding GM-CSF. This virus, administered intratumorally every 2 weeks for malignant melanoma, led to complete resolution in 47% of injected tumors and boosted systemic antitumor immunity leading to resolution of 9% of distant un infected visceral tumors [12] . Subsequent clinical studies have shown that responses are more frequent and more durable when T-Vec (and other OVs such as coxsackievirus A21) is combined with immune checkpoint For reprint orders, please contact: reprints@futuremedicine.com review Maroun, Muñoz-Alía, Ammayappan, Schulze, Peng & Russell future science group inhibitor antibody therapy [13] . Another significant milestone was the demonstration that a systemically administered oncolytic measles virus can target and destroy disseminated cancer in a human subject [11] .

engineering viruses: a diversity of platforms

As might be expected, proponents of a given OV are typically able to advance strong arguments to support their choice of platform, emphasizing unique features such as replication kinetics, genome plasticity, targetability, seroprevalence and stability that may lead to superior oncolysis. However, it is too early to determine which unique viral characteristics will be the critical drivers of clinical success for a given cancer type. Reverse genetics systems are available for virtually all virus families, and the rules of engagement for new virus creation are well established. In general, the most effective strategy is to combine rational design with evolution, allowing each engineered virus to mutate and fully adapt to its intended target cells after it has been rescued. Biosafety oversight is in place at all academic centers responsibly engaging in virus engineering activities, and it is now a relatively straightforward matter to generate and test new virus configurations using what now amounts to the world's best lego set. Viral gene and noncoding sequences can be modified in a variety of ways to add or eliminate functions and nonviral genes or noncoding regulatory elements, whether synthetic or naturally occurring, can be added into viral genomes to confer additional desirable properties. The overarching engineering goal for the oncolytic virotherapy field is to generate viruses that can be efficiently delivered to disseminated tumors in the body where they will spread and selectively kill both infected and uninfected tumor cells, without causing collateral damage and posing no risk of transmission to the population.

Delivery of Ovs

Clearly, the ability to infect the activated endothelial cells of tumor neovessels would be an attractive targeting property for OVs to enhance entry into the tumor parenchyma by releasing viral progeny on the ablumenal side of the blood vessel. Intravascular coagulation in the capillary could also be provoked by virusinfected endothelial cells reacting with clotting and inflammatory factors. Virus engineering strategies have been pursued to achieve this goal, for example, by displaying polypeptide ligands echistatin and urokinase plasminogen activator on the surface of measles virus to target integrin αVβ3 and UPaR endothelial cell surface receptors, respectively [68, 69] . Also, at least one OV (vesicular stomatitis virus VSV) has been shown naturally capable of infecting neovessel endothelium in implanted mouse tumors, but the study did highlight the potential toxicity of the approach, namely intravascular coagulation requiring heparin therapy for its prevention [70] .

• Limiting viral neutralization & clearance

Antibody neutralization has been shown to reduce the efficacy of systemically administered OVs such as measles, VSV and vaccinia in preclinical models [76] [77] [78] . When previous exposure has occurred, preformed antibody reduces the effectiveness of the initial treatment. However, the situation is typically far more problematic for second and subsequent doses, even for low seroprevalence viruses, because the first dose induces a powerful primary antiviral antibody response or boosts the pre-existing response [79, 80] . These troublesome responses can be constrained by coadministering immunosuppressive drugs such as cyclophosphamide. Cyclophosphamide administered concurrently with an OV has been shown to suppress or delay the development of humoral and cytotoxic antiviral T-cell responses [81] .
Another strategy for avoiding neutralization is to block the reticuloendothelial system with polyinosinic acid or with clodronate-loaded liposomes which poison or deplete splenic macrophages and Kupffer cells. This approach has been shown to slow the clearance of circulating virus particles that have been coated with antibodies or complement [86] [87] [88] .

• • Prodrug convertases

CD converts 5-flurocytosine (5-FC), an inert small molecule that is administered intravenously, to 5-flurouracil (5-FU), an antimetabolite that irreversibly inhibits thymidylate synthase and is an approved chemotherapeutic agent for a variety of cancers (anal, breast, colorectal, esophageal, stomach, pancreatic and skin) [29] . The argument supporting the CD/5-FC system is that local production of 5-FU in the OV-infected tumor will create a 5-FU concentration gradient that will expose tumor cells to a higher concentration of the drug compared with distant tissues, thereby ameliorating toxicity and enhancing the therapeutic index of the drug. However, 5-FU is freely diffusible so local production of the drug in a CD-positive tumor exposed to high concentrations of 5-FC can lead to systemic toxicity [140] . The dose of 5-FC must therefore be adjusted accordingly. CD has been incorporated into several OVs including adenoviruses, paramyxoviruses and poxviruses, but the one that has advanced ahead of all others in clinical testing is a C-type retrovirus, Toca-511, which is currently being evaluated in a Phase III clinical trial for the treatment of patients with malignant glioma [141] [142] [143] . Besides the TK and CD prodrug convertase systems, there are reports of OVs engineered to express the cyclophosphamide-activating protein CYP2B1, future science group www.futuremedicine.com the CPT11-activating secreted human intestinal carboxylesterase (shiCE) and the fludarabine phosphate activating purine nucleotide phosphorylase [144, 145] .

Safety

So far OVs have an exceptional safety record in the clinic with few serious adverse events and minimal mortality reported from human trials [166, 167] . However, since efficacy has also been limited, it remains possible that toxicity profiles may appear less favorable as doses are increased and as newer OVs or virus-drug combinations are advanced to clinical testing. Currently, there are over 40 active clinical trials ongoing using OVs alone or in combination with other therapies [168] . As the number of patients participating in trials grows so will the insight into rare adverse effects. The most common adverse effects reported in recent OV trials are fever and flu-like symptoms [169] . However, the potential for severe adverse reactions was much more clearly demonstrated in the earliest virotherapy trials undertaken in the 1950s and 1960s when deaths and severe adverse reactions were not infrequently recorded due to presumed OV replication in normal tissues, notably the brain, especially in immunocompromised cancer patients [4, 170, 171] .

• • Limiting pathogenicity

Two so-called suicide genes, TK and inducible caspase 9, are of interest in this regard. Both of these genes have been used as a safety switches to eliminate genetically modified T cells that were causing graft versus host disease in human clinical trials [184] [185] [186] . Viruses engineered to encode TK have also been controlled with ganciclovir therapy [187] , but this cannot be considered a reliable safety switch because of the ever-present risk of the emergence of viral quasispecies with TK inactivating mutations. The development of a universal and highly reliable safety switch remains one of the significant research challenges facing the field of oncolytic virotherapy.

• • Prodrug convertases

Unlike secreted toxins, prodrug convertases do not pose a risk of increasing OV virulence because their toxic potential is manifest only in the presence of an exogenously added prodrug. For this reason, the approach has been extensively studied. Perhaps the most well-known convertase-prodrug combination is HSV TK, used with ganciclovir. TK converts ganciclovir to ganciclovir monophosphate which is further processed intracellularly to ganciclovir triphosphate, a DNA synthesis chain terminator that kills dividing cells as they enter S phase [138] . A major weakness of the TK-ganciclovir system, aside from its inability to kill nondividing tumor cells, is that it has very limited bystander killing potential. This is because ganciclovir monophosphate is not released from the cell in which it is generated, so does not impact uninfected tumor cells unless they are connected to the infected cell via gap junctions through which it can pass [139] . Encoding connexin, a gap junction protein, in the OV genome can enhance the bystander killing effect of TK, but attention is shifting to other convertases, most notably CD.

Arming viruses with transgenes

• Secreted toxins, prodrug convertases and immunostimulatory proteins have been incorporated into OVs to increase treatment efficacy.
57 section matches

Abstract

Human Enterovirus 71 (EV71) commonly causes Hand, Foot and Mouth Disease in young children, and occasional occurrences of neurological complications can be fatal. In this study, a high-throughput cell-based screening on the serine/threonine kinase siRNA library was performed to identify potential antiviral agents against EV71 replication. Among the hits, Misshapen/NIKs-related kinase (MINK) was selected for detailed analysis due to its strong inhibitory profile and novelty. In the investigation of the stage at which MINK is involved in EV71 replication, virus RNA transfection in MINK siRNA-treated cells continued to cause virus inhibition despite bypassing the normal entry pathway, suggesting its involvement at the post-entry stage. We have also shown that viral RNA and protein expression level was significantly reduced upon MINK silencing, suggesting its involvement in viral protein synthesis which feeds into viral RNA replication process. Through proteomic analysis and infection inhibition assay, we found that the activation of MINK was triggered by early replication events, instead of the binding and entry of the virus. Proteomic analysis on the activation profile of p38 Mitogen-activated Protein Kinase (MAPK) indicated that the phosphorylation of p38 MAPK was stimulated by EV71 infection upon MINK activation. Luciferase reporter assay further revealed that the translation efficiency of the EV71 internal ribosomal entry site (IRES) was reduced after blocking the MINK/p38 MAPK pathway. Further investigation on the effect of MINK silencing on heterogeneous nuclear ribonucleoprotein A1 (hnRNP A1) localisation demonstrated that cytoplasmic relocalisation of hnRNP A1 upon EV71 infection may be facilitated via the MINK/p38 MAPK pathway which then positively regulates the translation of viral RNA transcripts. These novel findings hence suggest that MINK plays a functional role in the IRES-mediated translation of EV71 viral RNA and may provide a potential target for the development of specific antiviral strategies against EV71 infection.
Since its first isolation, human Enterovirus 71 (EV71) has been known to cause hand, foot and mouth disease in children, with some cases developing severe neurologic complications, leading to death. In the recent years, outbreaks within the Asia-Pacific region have caused significant deaths, making EV71 a major public health risk. Despite the growing threat from the spread of EV71 and increased research in this area, there are no clinically approved vaccines or antiviral drugs available against EV71. Cellular signalling and host kinases have been reported to play significant roles in the replication and propagation of many different pathogens. In this paper, we show that a serine/threonine kinase, Misshapen NIK-related kinase (MINK), plays a role in the replication of EV71 by stimulating the p38 mitogen activated protein kinase (p38-MAPK) pathway which in turns promotes the translation efficiency of EV71 viral protein synthesis. As the synthesis of viral proteins is crucial for the replication of the virus during infection, discovery of a crucial host kinase in this process may provide insights on the replication of EV71. With deeper understanding of the functions and regulation of MINK, this kinase may serve as a promising target for the development of antiviral therapy. MINK in the IRES-mediated Protein Translation of Human Enterovirus 71 PLOS Pathogens |

Silencing of MINK does not affect EV71 entry

Further experiments were performed to elucidate the involvement of MINK within the different stages of the EV71 replication processes (viral entry, viral RNA replication and viral protein synthesis). To assess the involvement of MINK in viral entry, viral RNA was extracted and transfected into cells which were pre-treated with MINK siRNA to bypass the normal viral entry processes ie. clathrin-mediated endocytosis for EV71 [26] . As such, infectious virus titre obtained from viral plaque assays would assist in the elucidation of the potential involvement of MINK in the viral entry stage. In this assay, a dose-dependent reduction in the virus yield was observed across the siRNA concentrations with a maximum reduction of~1.8 log at 45nM (Fig. 3A) , indicating that the silencing of MINK continued to cause virus inhibition despite bypassing the viral entry stage. This result suggested that MINK might play a more essential role at the post-entry stage.

Inhibition of the p38 MAPK signalling pathway affects viral protein synthesis

To further validate the involvement of p38 MAPK signalling pathway downstream of MINK in the synthesis of viral proteins, time-of-addition studies were conducted to identify the window period in the EV71 replication cycle when the p38 MAPK inhibitor, SB203580 exerts its antiviral effects. 50 μM of SB203580 was added at different time points before and after infection with EV71 (Fig. 6D ). All cell culture supernatants were collected for viral plaque assay at 12h post-infection. From the results in Fig. 6D , pre-treatment of cells with the p38 MAPK inhibitor for 2h prior to EV71 infection showed minimal inhibitory effect against viral infection. A significant reduction in EV71 titres was observed when SB203580 was added at 6h post-infection. At 10h post-infection, the antiviral activity of SB203580 was reduced. This suggested that the blocking of p38 MAPK signalling pathway inhibits EV71 replication during the early phase after viral entry, between 0h and 8h post-infection. A co-treatment assay was conducted to complement data from the pre-treatment assay to determine if SB203580 affected viral binding and entry into the cells. Co-treatment of EV71 with SB203580 failed to inhibit viral infection, further confirming the involvement of p38 MAPK signalling pathway in the post-entry stage in EV71 replication cycle.

Silencing of MINK and p38 MAPK inhibition reduced hnRNP A1 signals in the cytoplasm upon EV71 infection

To investigate if Mnk1 was involved in the mechanism elucidated in this study, we blocked Mnk1 kinase activity with a selective Mnk inhibitor (CGP57380) in RD cells and conducted Western blot analyses with phospho-eIF4E antibody to determine the effectivity of the drug. At 50μM, CGP57380 was effective in blocking the kinase activity of Mnk as indicated by the DMSO control, expressed as percentage. Error bars reflect the standard deviation of triplicate data sets. Transfected cells with the bicistronic construct without drug treatment (DMSO control) was used as positive control. Error bars represent standard deviation of triplicate data sets. Statistical analyses were performed using one-way ANOVA with Dunnett's test (Graphpad software). *P < 0.05, **P < 0.01 and *** P < 0.0001 versus 1.0% DMSO control.

Introduction

Human enterovirus 71 (EV71), a member of the Picornaviridae family and genus Enterovirus, is the major causative agent of hand-foot-and-mouth disease (HFMD). In recent years, EV71 has emerged as an important global health problem, causing significant deaths especially within the Asia-Pacific region [1] . Since its first isolation in 1969 in California [2] , outbreaks have been observed worldwide, affecting countries such as Singapore, Malaysia and Taiwan [3, 4, 5, 6] . EV71-associated HFMD often results in a higher risk of developing severe neurological complications and cardiopulmonary failure [7] which can be fatal. Despite the growing threat from the spread of EV71, there are no clinically approved vaccines or antiviral drugs available against EV71 to date [8] and treatments mainly aim to alleviate the symptoms [9] .
Given the small genome size of EV71, the virus depends on several cellular proteins and machineries in the host cell to complete its replication. In search for host cellular factors that play a role in EV71 replication, an understanding of the cellular proteins involved in different stages of viral replication would be useful for the identification of potential targets for the development of antiviral strategies. In this study, an immunofluorescence cell-based virus infection assay was set-up to screen the human serine/threonine kinase siRNA library using a set of validated small interfering RNAs (siRNAs) targeting the host serine/threonine kinases. Several candidate kinases that showed significant inhibition of EV71 replication upon gene knockdown were identified and among the hits, Misshapen/NIKs-related kinase (MINK), a novel sterile 20 (Ste20) family kinase, was chosen for further evaluations. MINK, also known as MAP4K6, is a germinal center kinase (GCK) from the Ste20 family of kinases that includes more than 30 serine/threonine kinases with catalytic domains that are homologous to the yeast Ste20 kinases. MINK is structurally similar to the Nck-interacting kinase (NIK) which has previously been proposed to link the protein tyrosine kinase signals to the activation of c-Jun N-terminus kinases (JNK) pathway via the SH2-SH3 domain of Nck [21] . As a member of the GCK class of kinases, MINK has an N-terminal kinase domain and a C-terminal regulatory domain. The intermediate domain consists of multiple proline rich motifs that are putative SRC homology 3 (SH3) binding sites [22] . The MINK1 gene encodes a polypeptide of 1312 amino acids and is expressed in most tissues in at least five alternatively spliced isoforms [22] . Studies on cells under environmental stress [23] revealed that MINK activates the JNK and p38 MAPK pathway, which are important signalling pathways involved in various cellular functions such as apoptosis, protein translation and cell differentiation [24] . Apart from cellular functions, p38 MAPK pathway has also been reported to play a role in the IRES-mediated viral protein translation of Encephalomyocarditis virus (EMCV) viral RNA [25] . In this study, we revealed the involvement of MINK in EV71 replication and further elucidated the mechanism through which MINK regulates the synthesis of EV71 viral polyprotein upon viral infection.

MINK plays an essential role in EV71 replication

In view of the higher levels of virus inhibition upon the silencing of MINK and its unknown function in virus replication, MINK was selected for further investigation. Western blotting was carried out to verify the gene knockdown efficiency. Dose-dependent reduction in the protein expression level of MINK was observed upon MINK siRNA treatment, suggesting that the range of siRNA concentration used in this study was effective in silencing the MINK gene ( Fig. 2C panel i and Fig. 2D ). This was further verified by the scrambled siRNA treatment, as Human serine/threonine kinase siRNA library screen. Effect of gene knockdown by siRNA on EV71 replication analysed from primary screen. 40% reduction in viral antigen positive cells was considered as the acceptable level of virus inhibition and positive hits are genes which resulted in a percentage of viral antigen positive cells of less than 60% upon the knockdown of these genes. As such, 6 genes have been identified as positive hits from the primary screen. First 6 bars represent siRNA controls used while the other bars represent the host serine/threonine kinases targeted in the screening. siRNA controls utilised included non-targeting siRNAs as well as siRNAs targeting several housekeeping genes. Values obtained in the graph were normalised against the mean of the transfection control (EV71-infected cells treated with only the transfection reagent). there was no reduction in MINK protein levels observed across the siRNA concentration range used in the study (Fig. 2C panel ii, and Fig. 2D ). To further validate the specificity of the MINK SMARTpool siRNAs, deconvolution assay was performed with 45nM concentrations of each specific individual siRNA of the SMARTpool (four specific siRNAs). This approach would help to ensure that inhibitory effects of the targeting siRNAs on EV71 infection observed in the secondary assay was specific and not due to off-target gene effects. Before a viral plaque assay was performed to determine the virus titre upon gene knockdown using the individual siRNAs within the SMARTpool, gene knockdown efficiency of each individual siRNAs was assessed. From the Western blot analysis, it was observed that all four individual siRNAs directed against MINK were effective in reducing MINK protein levels at a concentration of 45nM ( Fig. 2E and Fig. 2F ). Coinciding with the knockdown efficiency of individual siRNA, viral plaque assay results showed that all four individual siRNAs resulted in significant reduction in virus titre of at least 1.3 log at 45nM concentration (Fig. 2G) . Taken together, these findings demonstrated that the inhibition on EV71 propagation was a result of the targeted siRNA knockdown of MINK and thus, MINK is essential for the replication of EV71.

Discussion

Apart from factor eIF4E, other canonical translation factor such as eIF2α has also been reported to play a role in Picornavirus replication [54] . Factor eIF2α is a 36kDa protein that contains a serine residue (Ser-51) which can be phosphorylated under nutrient deprivation or cellular stresses such as virus infection or heat-shock. GCN2, PKR, PERK and HRI have been shown to phosphorylated eIF2α in response to amino acid starvation, double-stranded RNA, protein misfolding at the endoplasmic reticulum and the absence of HEME, respectively [55] . Phosphorylation of eIF2α inhibits the GDP-GTP recycling catalysed by eIF2B, hindering the generation of the ternary complex Met-tRNAi-eiF2-GTP and binding of this complex to the 40S ribosome to initiate translation [56] . Although eIF2α is required for both cap-dependent and IRES-mediated protein translation, studies have shown that some viral IRES elements can translate independent of phosphorylation of eIF2α [57, 58] . Consistent with studies on enteroviruses [32] , we observed an increasing level of phosphorylated eIF2α at late times post-infection (S3A and S3B Fig). Published studies on poliovirus have also demonstrated that resistance to eIF2α phosphorylation increases as enteroviral infection progresses due to the cleavage of initiation factor eIF5B by the viral 3C protease. As such, the induction of eIF2α phosphorylation at the late time-points of EV71 infection may also serve to promote the viral protein synthesis indirectly by suppressing cellular cap-dependent protein synthesis [59] . Although p38 MAPK signalling has not been implicated in the phosphorylation of eIF2α and significant reduction in cap-dependent protein translation was not observed in our luciferase data, we have conducted brief experiments to investigate the relationship between MINK expression and the phosphorylation of eIF2α (S3C and S3D Fig) . From the suppression of eIF2α phosphorylation upon the silencing of MINK, it is tempting to speculate that the phosphorylation of eIF2α may be a minor side effect of the activation of MINK that serve to promote EV71 protein synthesis. Future downstream studies have to be performed to elucidate the role and involvement of eIF2α in EV71 replication in relation to MINK. Nonetheless, our findings on these canonical translation initiation factors suggested that the increased EV71 IRES translation efficiency observed in our study might have resulted from the activation of ITAFs downstream of MINK/ p38 MAPK signalling instead of the phosphorylation status of the eIFs.

Cell line and viruses

siRNA Dharmacon siGENOME Human SMARTpool custom siRNA library targeting human serine/ threonine kinases was obtained from Dharmacon RNA Technologies (Thermo Scientific, Dharmacon RTF # H-004405). The library contains a total of 47 siRNA cocktails with each cocktail consisting of 4 siRNA sequences targeting a specific gene and resuspended at a concentration of 2μM in 96-well plates. The separate and individual siRNA pools that were used to validate the hits identified from the screen was also obtained from Dharmacon (siGENOME

Introduction

EV71 is a small (33-35nm) , single-stranded, positive-sense, non-enveloped RNA virus with a viral genome of approximately 7.5kb. The virions consist of an icosahedral capsid of 60 protomers surrounding viral genomic RNA [10] that contains a single open reading frame (ORF) flanked by the 5' untranslated region (UTR) and the 3' UTR. The ORF encodes four structural proteins (VP1, VP2, VP3 and VP4) that make up the viral capsid and seven non-structural proteins (2A, 2B, 2C, 3A, 3B, 3C and 3D) which are involved in viral replication. Viral 2A and 3C proteases are involved in the cleavage of the polyprotein precursor to release the mature viral proteins while viral protein 3D is the RNA-dependent RNA polymerase (RdRp) that plays a major role in the synthesis of negative-and positive-sense viral RNA [11] . Upon EV71 infection, the viral genome is translated into the viral polyprotein and the 3D protein participates in the transcription of the positive-sense genomic viral RNA into the complementary negative-sense viral RNA, which serves as a template for the synthesis of more positive-sense genomic viral RNA. The genomic RNA is then translated into more viral polyproteins in a cap-independent manner and the polyproteins are subsequently processed into the structural capsid proteins and non-structural proteins [11] . The 5' UTR of the EV71 genomic RNA contains a cloverleaf structure involved in viral RNA replication and an internal ribosomal entry site (IRES) which directs viral protein translation in a cap-independent manner [12, 13] . The picornavirus IRES have been classified into three types based on its structure and enteroviruses (and rhinoviruses) have the type 1 IRES which requires certain eukaryotic initiation factors (eIFs) and IRES-specific transacting factors (ITAFs) to initiate viral protein translation. In contrast to cellular capdependent translation, the host 40S ribosomal subunit is recruited at the IRES without the need for eIF4E to initiate viral polyprotein translation [14] . A number of ITAFs have been identified to interact with picornavirus IRES and mediate translation initiation of the viral polyprotein. These ITAFs include polypyrimidine tract-binding protein (PTB) [15, 16, 17] , heterogeneous nuclear ribonucleoprotein E (hnRNP E) [18] , far-upstream element-binding protein 1 (FBP1) and FBP2 [19] . Among these ITAFs, hnRNP A1 [20] ,and FBPs [19] have been reported to interact with EV71 IRES.

Development of a screening assay for EV71 replication based on indirect immunofluorescence

A screening assay was previously developed to screen for host factors involved in EV71 replication using targeting siRNA [26] . In this screen, a similar screening approach was adopted based on immunofluorescence assay to detect EV71 structural protein expression as an indicator of successful EV71 infection and replication. Positive control wells containing EV71infected cells without siRNA treatment had a mean of 51.600% antigen positive cells with a standard deviation of 5.407%. Mock-infected cells were used as negative controls to verify the specificity of the antibody. In addition, the data were analysed by applying Z-score statistics and a Z' factor of 0.673 was obtained from the primary screen, indicating that the screening platform was sufficiently robust and was suitable for the high-throughput screening of the human serine/threonine kinase siRNA library.

MINK plays an essential role in EV71 replication

A human serine/threonine kinase siRNA library that targets 47 serine/threonine kinases (S1 Table) was utilised in the primary screen in search for human serine/threonine kinases involved in EV71 replication. Using the criteria of 40% inhibition to identify positive hits in the primary screen, 6 serine/threonine kinases were identified as positive hits (Fig. 1) . The top three targets identified, PAK1, MINK and MAP4K2, were first analysed. PAK1 was observed to cause the highest level of virus inhibition but closer analysis led to the removal of this target from further downstream studies due in part to the large standard error in the results obtained and lower cell density observed in all the PAK1 replicate wells. MINK and MAP4K2 were identified as putative targets crucial to EV71 replication and validation of their involvement was carried out using both immunofluorescence assay and viral plaque assay.
Increasing concentrations of siRNA resulted in a reduction of immunofluorescencedetectable EV71 replication for cells treated with siRNA against either MINK or MAP4K2 ( Fig. 2A ). This coincided with the dose-dependent reductions up to 1.5 log for both siRNAs at 45nM in infectious virus released from the cells as indicated by viral plaque assays (Fig. 2B ). As the siRNA targeting cyclophilin B has been utilised previously in the primary screen as an siRNA control, it was also included in the secondary assay to ensure that the effect of infectious virus titre reduction was not due to off-target effects. Furthermore, minimal cellular cytotoxicity was observed across the concentrations of both MINK and MAP4K2 siRNAs (Fig. 2B) , indicating that the inhibition of EV71 replication was not due to the cytotoxic effects of the siRNAs at the range of concentrations used.

MINK is essential for the replication of other human enteroviruses

To investigate if MINK plays a role in other human enteroviruses as well, MINK siRNA-treated cells were infected with various human enteroviruses: a different strain of EV71 (EV71 strain 41), Coxsackievirus A6 (CA6) and Echovirus 7 at MOI 1. The dose-dependent reduction in infectious virus titres upon the siRNA knockdown of MINK was reproduced with all three viruses as demonstrated in Fig. 2H . As shown in Fig. 2H , the knockdown of MINK resulted in the reduction of infectious virus titre by approximately 2.3 log units. 1.5 log unit and 1.5 log unit respectively for EV71 strain 41, Echovirus 7 and CA6 relative to the scrambled siRNA controls (Scr). This suggested that the involvement of MINK is not restricted to EV71, but may extend to other human enteroviruses as well.

Silencing of MINK does not affect EV71 entry

MINK plays an essential role in EV71 viral protein synthesis and viral RNA synthesis
To examine the involvement of MINK in the post-entry stages of EV71 replication, viral RNA synthesis of EV71 was determined by quantitative RT-PCR on viral RNA samples extracted from infected RD cells pre-treated with either MINK siRNA (M) or scrambled siRNA (S) control at 25nM (M25 and S25) and 45nM (M45 and S45). A background control measuring the viral RNA levels at 0h post-infection was included to account for the background viral RNA resulted from virus entry and the binding of residual virions on the cell surfaces. The infected Silencing of MINK significantly reduced EV71 replication in a siRNA concentration-dependent manner. (A) siRNA-treated EV71 infected cells were fixed at the same time-points and intracellular viruses were detected by immunofluorescence assay. Immunofluorescence detection of EV71 VP2 proteins (green) with the nuclei stained with DAPI (blue) is shown. The images were taken at 10X magnification. Cells in both the negative and transfection controls were not infected with EV71 while cells in the positive control were infected with EV71 in the absence of siRNA. (B) Cell viability of siRNA-treated cells was measured in relation to untreated cells using alamarBlue assay after 72h incubation. Virus titres in the supernatant of siRNA-treated cells were analysed via viral plaque assay. Error bars represent standard deviation (SD) of triplicate data and values obtained were normalised against the transfection control. Statistical analyses were performed using one-way ANOVA and Dunnett's test (Graphpad software) against untreated control. *P <0.05 (n = 3), **P <0.01 (n = 3). (C) Verification of gene knockdown efficiency of MINK siRNA SMARTpool at concentrations ranging from 0nM to 45nM. Western blot analysis was performed to detect protein expression levels of MINK, with β-actin as the loading control. Parallel transfection of scrambled siRNA served as a cells pre-treated with the siRNA were harvested at 8h and 10h post-infection to examine the relative amount of viral RNA. Fold change in RNA level for all samples was calculated relative to the RNA level in the transfection control (PTC/ 0nM) at 0h post-infection. Comparison of RNA level was made between the samples treated with MINK siRNA (M25 and M45) or scrambled siRNA (S25 and S45) and their PTC at each time-point. Results showed increase in the level of viral RNA at 8h and 10h post-infection, relative to 0h background control, indicating that there was viral RNA replication upon siRNA treatment. The viral RNA level in MINK siRNA-treated cells was significantly lower than that in the transfection control (PTC) at both 8h and 10h post-infection (Fig. 3B) , with fold reductions of 6.8 and 18.4 at 8h and 10h post-infection at 45nM concentration, respectively. On the contrary, there was no significant change in the viral RNA level in the samples treated with the scrambled siRNA control at both concentrations compared to PTC. These results indicated that the silencing of MINK has inhibitory effects on the production of viral RNA.
Since the production of viral proteins precedes and is essential for the synthesis of viral RNA, the influence of MINK silencing on EV71 replication at the level of translation was determined. A time course study was first conducted to identify the time period during which EV71 viral protein expression occurs ( Fig. 3C and 3D ). Across a time course of 12h after EV71 infection, greatest increase in the EV71 structural viral protein expression, VP0 (~36kDa) and VP2 (~28kDa), was observed between 4 and 8h post-infection ( Fig. 3C and 3D ). As such, 8h postinfection was selected as the time-point for further analysis on the role of MINK in viral protein synthesis. At 8h post-infection, a dose-dependent reduction in the VP0 and VP2 was observed in EV71-infected cells pre-treated with MINK siRNA (

Silencing of MINK does not block virus release

To further confirm the stage of involvement of MINK in the replication of EV71, intracellular and extracellular EV71 virions were quantified at 12h post-infection to determine whether MINK plays a role in viral packaging and release. Although a significant reduction of~1.5 log was observed in the extracellular virions upon the siRNA knockdown of MINK at 45nM, a significant reduction in virus titre was also observed in the intracellular virions (Fig. 3G) . Hence, we concluded that the reduction in the amount of virus released was not due to a blockage in knockdown control. MINK protein expression was observed to decrease in a dose-dependent manner across siRNA concentration. (D) Band intensity of MINK gene knockdown verification. The band intensities representing MINK protein expression level were quantitated with reference to actin control bands (for each individual concentration) and PTC. The intensities of protein bands were quantitated using ImageJ Gel Analysis program. (E) Verification of gene knockdown efficiency of individual siRNA within the siRNA SMARTpool directed against MINK at 45nM. Western blot analysis was performed to detect protein expression levels of MINK, with β-actin as the loading control. (F) Band intensity of MINK gene knockdown verification in deconvolution assay. The band intensities representing MINK protein expression level were quantitated with reference to actin control bands (for each individual siRNA) and PTC. (G) Virus titres in the supernatant of cells treated with individual siRNAs within siRNA SMARTpool were analysed via viral plaque assay. Error bars represent standard deviation (SD) of triplicate data. Statistical analyses were performed using one-way ANOVA and Dunnett's test (Graphpad software) against untreated control. ***P <0.0001 (n = 3). (H) Virus titres of other human enteroviruses (Echovirus 7, Coxsackievirus A6 and EV71 strain 41) in the supernatant of siRNA-treated cells were analysed via viral plaque assay. Error bars represent standard deviation (SD) of triplicate data. Statistical analyses were performed using one-way ANOVA and Dunnett's test (Graphpad software) against scrambled control (Scr). *P < 0.05, **P < 0.01 and ***P < 0.0001 (n = 3) versus scrambled control. MINK plays an essential role in EV71 viral protein synthesis. (A) EV71 viral RNA was transfected into RD cells pre-treated with MINK siRNA and supernatant was harvested from cells at 12h post-infection (hpi) for viral plaque assay. Silencing of MINK with targeting siRNA continued to cause inhibition of virus replication. Statistical analysis was performed using one-way ANOVA with Dunnett's test (Graphpad software). *** P < 0.0001 (n = 3) versus untreated control (0nM). (B) EV71 RNA synthesis was sensitive to silencing efficiencies of MINK. Quantitative RT-PCR assay revealed significant reduction in levels of EV71 RNA across increasing siRNA concentration in MINK siRNA-treated cells. Total RNA was extracted for all samples at 0, 8 and 10hpi and EV71 RNA levels were measured. C T values were normalised against actin and relative quantification of viral RNA level was determined. The ΔΔCt data were calculated from three independent experiments and error bars represent standard deviation for triplicate data sets. Fold difference of viral RNA for all samples was calculated relative to the RNA level in the transfection control (PTC) at 0hpi. Statistical analyses were carried out using one-way ANOVA with Dunnett's test (Graphpad software). *P<0.05 and *** P < 0.0001 (n = 3) vs the respective PTC at each time-point. (C) Time course study of EV71 structural protein expression via Western blot analysis. Upper band (36kDa) represents VP0 while lower band (28 kDa) represents VP2. β-actin was used as the the virus release process upon the silencing of MINK, but was due to a decrease in the total production of virus particles.

Replication of EV71 triggers MINK phosphorylation

As a MAP kinase kinase kinase kinase (MAP4K), MINK is activated upstream in MAPK pathways and thus we hypothesised that the early events in EV71 infection could be responsible for the activation of MINK. To investigate if virus binding and entry triggered the phosphorylation of MINK, Western blot analysis on phospho-MINK was conducted after the transfection of viral RNA into cells to bypass the normal entry processes of EV71. Since phospho-MINK antibodies are not available commercially, a phosphate-binding tag (Phos-Tag) [27] was used to separate the phosphorylated proteins from the unphosphorylated proteins. 6h and 8h were selected as the time-points for harvest of cell lysates due to the significant increase in phospho-MINK levels at these time-points after infection (S1 Fig). As shown in Fig. 4A and Fig. 4B , cells transfected with viral RNA displayed similar phospho-MINK levels at 6h and 8h as the infection control (EV71-infected), suggesting that initial binding and entry processes of the virus was not required for the activation of MINK upon EV71 infection.
To investigate if virus binding to cellular SCARB2 triggered the activation of MINK, EV71infected RD cells pre-treated with the anti-SCARB2 antibody were lysed at 6h after addition of virus for Western blot analysis. Our results indicated that inhibition of the virus binding to SCARB2 with increasing concentration of the antibody did not reduce the phospho-MINK levels in the cells as the phospho-MINK levels in cells treated with 25 and 50μg/mL of anti-SCARB2 antibody showed similar level as that in cells treated with 50μg/mL of control IgG antibody ( Fig. 4D and 4E ). As such, virus binding was unlikely to be the triggering event of the phosphorylation of MINK. Together, the activation profile of MINK (S1 Fig) and the entry assays suggested that the phosphorylation of MINK was stimulated post-entry, in the early phase of viral replication which occurs during the 6h period after addition of virus. loading control. (D) Band intensity of VP0 and VP2 in time course study. The band intensities representing VP0 and VP2 protein expression level were quantitated with reference to actin control bands (for each time-point) and 0hpi using ImageJ Gel Analysis program. (E) Viral protein expression levels upon the silencing of MINK. VP0 and VP2 viral protein expression was observed to decrease with increasing concentration of siRNA targeting MINK. (F) Band intensities of VP0 and VP2 upon siRNA knockdown of MINK. The band intensities representing VP0 and VP2 protein expression level were quantitated with reference to actin control bands (for each siRNA concentration) and 0nM using ImageJ Gel Analysis program. (G) Extracellular and Intracellular virion levels upon the silencing of MINK. Extracellular EV71 virions in the supernatant and intracellular virus particles were harvested separately at 12hpi for viral plaque assay to assess the effect of siRNA knockdown of MINK on virus packaging and release. Silencing of MINK resulted in significant reduction in both intracellular and extracellular virions. Statistical analysis was performed using one-way ANOVA with Dunnett's test (Graphpad software). **P < 0.01 and *** P < 0.0001 (n = 3) versus untreated control (0nM). Phosphorylation of MINK is triggered post-entry by early replication events. Phos-tag acrylamide binds phosphorylated proteins and retards their migration to separate the phosphorylated proteins from their unphosphorylated counterparts. Total MINK antibody was used to detect both phosphorylated (upper bands) and unphosphorylated MINK (lower bands). β-actin was used as a loading control. (A) Viral RNA was transfected into cells and cell lysates were harvested at indicated time-points to assess the phospho-MINK levels. Phospho-MINK levels in RNA-transfected cells were comparable to the infection control at the same time-points. (B) The band intensities representing MINK phosphorylation level were quantitated with reference to actin control bands (for each time-point) and 0h using ImageJ Gel Analysis program. (C) Virus titres in the supernatant of cells treated with the anti-SCARB2 and anti-IgG antibodies were analysed via viral plaque assay. Blocking SCARB2 receptors with increasing concentration of SCARB2 antibody resulted in a significant reduction in virus titres. Error bars represent standard deviation (SD) of triplicate data. Statistical analyses were performed using oneway ANOVA and Dunnett's test (Graphpad software) against untreated control. ***P <0.0001 (n = 3) (D) Blocking SCARB2 receptors with increasing concentration of SCARB2 antibody did not affect the phosphorylation of MINK in cells at 6h after addition of virus. (E) The band intensities representing MINK phosphorylation level were quantitated with reference to actin control bands (for each concentration) and 0μg/mL using ImageJ Gel Analysis program.

EV71 infection activates p38 MAPK in RD cells

After determining the triggering event of MINK upon EV71 infection, we next investigated the mechanism of action of MINK on EV71 viral protein synthesis. It has been reported that MINK activates the p38 MAPK pathway [23] , a signalling pathway that has also been shown to play a role in the replication of Encephalomyelitis virus (EMCV), a member of the Picornaviridae family [25] . As such, we examined the activation profile of p38 MAPK upon EV71 infection to assess whether the p38 MAPK signalling pathway is activated during EV71 replication. As serum has also been reported to induce phosphorylation of certain proteins [29] , fecal calf serum (FCS) was removed from the virus stock and growth media in the course of this experiment to reduce the additional activation of p38 MAPK by the serum. Cell lysates were analysed at indicated time-points post-infection for 12h by Western blotting to examine the changes in the phosphorylation levels of p38 MAPK (phospho-p38). Constant and basal phosphorylation of p38 MAPK was observed in mock-infected cells throughout the 12h time course (Fig. 5B and 5D). In contrast, the EV71-infected cells showed an increase in the phosphorylation level of p38 MAPK between 6 to 8h post-infection ( Fig. 5A and 5D ), followed by a subsequent decrease from 8 to 12h post-infection. To demonstrate the dependency of p38 MAPK phosphorylation on EV71 replication, we also examined the phospho-p38 MAPK profile in RD cells infected with UV-inactivated EV71 ( Fig. 5C and 5D ). Similar to the mock-infected control, cells exposed to UV-inactivated EV71 showed constant phosphorylation level of p38 MAPK throughout the 12h time course, indicating that attachment of the virions to cell surface receptors or virus entry process were not sufficient to trigger the phosphorylation of p38 MAPK. Total p38 MAPK was also assessed to ensure that the changes in phospho-p38 MAPK levels were not due to differences in p38 MAPK expression levels. Phospho-p38 MAPK levels at 0h post-infection appears higher in Fig. 5A in EV71-infected samples than that in the mockinfected samples (Fig. 5B ) and samples exposed to UV-inactivated EV71 (Fig. 5C ) probably due to more total proteins loaded as seen from the total p38 and β-actin levels. However, it is evident in the trend of p38 MAPK activation profile that phospho-p38 MAPK levels were significantly increased upon EV71 infection (Fig. 5D) . Hence, these results suggested that activation of p38 MAPK signalling pathway requires the active replication of EV71.

MINK contributes to the activation of p38 MAPK during EV71 infection

To establish a link between the activation of MINK and p38 MAPK phosphorylation upon viral infection, cells were pre-treated with either scrambled MINK or MINK siRNA prior to EV71 infection and lysed at 8h post-infection for Western blot analysis. Efficacy of the MINK siRNA was demonstrated with a dose-dependent reduction in MINK protein levels upon knockdown ( Fig. 5E and 5F ). Dose-dependent reduction in phospho-p38 MAPK protein levels was also observed with increasing MINK siRNA concentration (25nM and 45nM) in both EV71-infected (Fig. 5E left panel and 5G ) and mock-infected samples (Fig. 5E right panel and 5G ). In contrast, increasing the concentration of scrambled siRNA control (25nM and 45nM) did not affect the phosphorylation levels of p38 MAPK ( Fig. 5E and 5G) , suggesting a correlation between MINK expression levels and p38 MAPK phosphorylation levels. As such, these results confirmed that MINK plays a role in the downstream triggering of p38 MAPK phosphorylation during EV71 replication.

The p38 MAPK signalling pathway is essential for EV71 replication in RD cells

Having confirmed the effectiveness of SB203580 in inhibiting the phosphorylation of p38 MAPK, RD cells were post-treated with SB203580 and the production of progeny virus in the culture supernatant was assessed via viral plaque assay. A significant dose-dependent reduction in the progeny virus production was observed after treatment with SB203580 with a 1.2-, 2.2and 3.3 log reduction at 25 μM, 75 μM and 100 μM, respectively (Fig. 6C ). In addition, cell viability of SB203580-treated cells was assessed to rule out the possibility of reduced viral growth due to cytotoxicity after treatment with SB203580. Our results from the alamarBlue cytotoxicity assay indicated that the concentration range of SB203580 used in this study did not lead to significant reductions in cell viability and hence, the dose-dependent reduction of EV71 virus titre by SB203580 was not complicated by its cytotoxic effects (Fig. 6C ). Together, these results indicated that blockage of the p38 MAPK signalling pathway can significantly reduce viral growth and hence, p38 MAPK pathway is essential for EV71 propagation.

Inhibition of the p38 MAPK signalling pathway affects viral protein synthesis

To reaffirm our hypothesis that MINK affects viral protein synthesis via p38 MAPK signalling pathway, the effect of p38 MAPK inhibition on viral RNA production, protein synthesis and viral release was investigated. In the analysis of viral RNA replication in SB203580-treated infected cells, fold change in RNA level for all samples was calculated relative to that in the DMSO control at 0h post-infection. Comparison of the relative RNA level was made between the samples treated the p38 MAPK inhibitor (SB203580) and the respective DMSO control at each time-point. Similar to what was observed for the siRNA knockdown of MINK (Fig. 3B) , there was increase in viral RNA level between time-points and viral RNA level in SB203580treated cells was significantly lower than that in the DMSO control at both 8h and 10h post-infection, with fold reductions of 12.4 and 9.0 at 8h and 10h post-infection at 100μM concentration, respectively (Fig. 6E) . These results indicated that the blocking of p38 MAPK pathway also has inhibitory effects on the production of viral RNA. Significant reduction in the Fig. 6F and 6G ) and non-structural protein levels (3D protein, S2D and S2E Fig) were also observed with increasing concentrations of SB203580. The extent of the reduction in 3D protein level upon SB203580 treatment corresponded to the decrease in the viral RNA levels. Hence, we have shown that p38 MAPK signalling pathway is involved in the viral protein synthesis stage.
As p38 MAPK signalling has been shown to play a critical role in the viral progeny release of coxsackievirus B3 (CVB3) [31] , we were interested to know if p38 MAPK was also involved in the viral release of EV71 using the p38 MAPK inhibitor (SB203580). Intracellular and extracellular EV71 virions were quantified at 12h post-infection to determine whether p38 MAPK plays a role in viral packaging and release. Although a significant reduction of~1.9 log was observed in the extracellular virions upon the inhibition of p38 MAPK at 50μM, a significant reduction in virus titre was also observed in the intracellular virions (Fig. 6H) . Hence, we concluded that in accordance to what was observed with the siRNA knockdown of MINK, the reduction in the amount of virus released was not due to a blockage in the virus release process upon the inhibition of p38 MAPK signalling, but was due to a decrease in the total production of virus particles.
Together, these results coincided with our previous observations with the siRNA knockdown of MINK (Fig. 3) and supported our hypothesis that p38 MAPK signalling pathway is very likely to be involved downstream of MINK in regulating EV71 viral protein synthesis.

MINK/p38 MAPK signalling positively regulates EV71 viral protein translation efficiency

Translation initiation of the uncapped EV71 viral RNA is known to be mediated by a cap-independent mechanism which involves the IRES situated in the 5' UTR of the viral genome [32] . Since the silencing of MINK reduced viral protein synthesis, we want to investigate if MINK/ p38 MAPK signalling was involved in the regulation of IRES efficiency during viral protein synthesis. To determine if MINK/p38 MAPK is involved in IRES-mediated translation of 0.0001 (n = 3) versus 1.0% DMSO control. (D) RD cells were treated with 50uM SB203580 (p38 MAPK inhibitor) at different time points before and after infection in time-of-addition assay. Cell supernatants were harvested at 12hpi for quantification via viral plaque assays. Time-of-addition assay indicates that SB203580 acts between 2hpi and 10hpi to inhibit EV71 replication. In the co-treatment assay, SB203580 was added with the virus and no significant inhibition of EV71 infection was observed. Statistical analyses were performed using one-way ANOVA and Dunnett's test (Graphpad software) against untreated control **P < 0.01 (n = 3), *** P < 0.0001 (n = 3) versus 0.5% DMSO control. (E) EV71 RNA synthesis was sensitive to SB203580 treatment. Quantitative RT-PCR assay revealed significant reduction in levels of EV71 RNA across increasing SB203580 concentration. Total RNA was extracted for all samples at 0, 8 and 10hpi and EV71 RNA levels were measured. C T values were normalised against actin and relative quantification of viral RNA level was determined. The ΔΔCt data were calculated from three independent experiments and error bars represent standard deviation for triplicate data sets. Fold difference of viral RNA for all samples was calculated relative to the RNA level in the DMSO control at 0hpi. Statistical analyses were carried out using one-way ANOVA with Dunnett's test (Graphpad software). *P <0.05, **P <0.01 and *** P < 0.0001 (n = 3) vs the respective 1.0% DMSO control at each time-point. (F) Viral protein expression levels upon SB203580 treatment. EV71-infected RD cells were treated with SB203580 and cell lysates were harvested for Western blotting at 8h post-treatment. VP0 and VP2 viral protein expression was observed to decrease with increasing concentration of the p38 MAPK inhibitor. (G) Band intensities of VP0 and VP2 upon SB203580 treatment. The band intensities representing VP0 and VP2 protein expression level were quantitated with reference to actin control bands (for each concentration) and DMSO control using ImageJ Gel Analysis program. (H) Extracellular and intracellular virion levels upon p38 MAPK inhibition. Extracellular EV71 virions in the supernatant and intracellular virus particles were harvested separately at 12hpi for viral plaque assay to assess the effect of SB203580 treatment on virus packaging and release. p38 MAPK inhibition resulted in significant reduction in both intracellular and extracellular virions. Statistical analysis was performed using one-way ANOVA with Dunnett's test (Graphpad software). **P <0.01, *** P < 0.0001 (n = 3) versus 1.0% DMSO control.
doi:10.1371/journal.ppat.1004686.g006 EV71 transcripts, a bicistronic luciferase reporter construct containing the EV71 IRES and two luciferase reporter genes (firefly luciferase and Renilla luciferase, Fig. 7A ) was transfected into cells. The expression of Renilla luciferase (RLuc) is dependent on the cap-dependent mechanism, while the translation of the firefly luciferase (FLuc) is IRES-dependent. The translation efficiency directed by the 5' UTR of EV71 was determined by comparing the level of FLuc with the level of RLuc after the transfection of the non-replicating bicistronic luciferase reporter construct into RD cells that had been pre-treated with MINK siRNA. 0.5mg/ml of amantadine was added as a negative control (Amantadine-treated) to inhibit the EV71-IRES [33] while untreated cells serve as a positive control. MINK siRNA-treated cells showed a dose-dependent reduction in the EV71 IRES activity (FLuc level) and the relative translation efficiency of the IRES with significant decrease observed at 15nM (Fig. 7B ). Cells treated with non-targeting siRNA (scrambled siRNA) showed minimal effect on the EV71 IRES activity as the relative translation efficiency remained relatively constant across the concentrations of scrambled siRNA with a non-significant decrease of 0.2 RLU at 45nM. In addition, a comparison of the RLuc levels of the scrambled control and MINK siRNA-treated cells indicated that the siRNA knockdown of MINK had minimal effect on cap-dependent translation.
To further confirm our hypothesis that p38 MAPK is involved in the translational regulation of viral transcripts downstream of MINK activation, RD cells were transfected with the bicistronic luciferase reporter construct before they were treated with different concentrations of SB203580. Results showed significant dose-dependent reduction in the relative translation efficiency of EV71 IRES in cells treated with SB203580 ( Fig. 7C ) relative to the DMSO control without affecting the cap-dependent translation. Thus, these data indicated that MINK/p38 MAPK is involved in the positive regulation of IRES-mediated translation of EV71 RNA.

Silencing of MINK and p38 MAPK inhibition reduced hnRNP A1 signals in the cytoplasm upon EV71 infection

Since eIF4E was unlikely to be the effector of the MINK/p38 MAPK pathway in our study, we hypothesized that the MINK/p38 MAPK signalling pathway might activate an IRES transacting factor (ITAF) downstream, which thus resulted in the positive regulation of the EV71 IRES translation efficiency. Heterogenous nuclear ribonucleoprotein A1 (hnRNP A1) is predominantly a nuclear protein but shuttles back and forth between the nucleus and cytoplasm. It has been reported that hnRNP A1 act as an ITAF that relocalises in the cytoplasm where it interacts with the EV71 IRES upon infection, promoting its translation efficiency [20] . As previous study has also shown that p38 MAPK signalling is implicated in the cytoplasmic accumulation of hnRNP A1 in uninfected cells [35] , we were interested to know whether MINK/p38 MAPK signalling modulate the EV71 IRES activity by altering the subcellular localisation of hnRNP A1. Fig. 8A shows the immunofluorescence staining of hnRNP A1 in EV71-infected cells at 8h post-infection. The degree of colocalisation between the hnRNP A1 protein (stained with rhodamine) and the cell nucleus (stained with DAPI) was quantified using Manders RD cells were pre-treated with MINK or scrambled siRNA. Three days after transfection, the bicistronic construct was then transfected into the cells. Luciferase activity was measured 24h after transfection. Amantadine, an inhibitor of EV71 IRES [33] , was added to untreated cells to serve as negative control for IRES activity. Untreated cells that were transfected with the bicistronic construct were used as positive control. The FLuc/RLuc ratio for each sample were normalised to the FLuc/RLuc ratio of untreated control. Dose-dependent reduction in relative translation efficiency of the IRES was observed in MINK siRNA-treated cells. Error bars represent standard deviation of triplicate data sets. Statistical analyses were performed using one-way ANOVA with Dunnett's test (Graphpad software). *P < 0.05, **P < 0.01 and *** P < 0.0001 versus untreated control. (C) Effect of p38 MAPK inhibition on EV71 IRES activity. Relative translation efficiency was determined as the ratio of FLuc to RLuc for each sample and the FLuc/ RLuc ratio for each sample were normalised to the FLuc/RLuc ratio of coefficient [36] . The level of the hnRNP A1 in the nucleus increased significantly upon the siRNA-knockdown of MINK (74.8% colocalisation, Fig. 8A xiii-xvi) compared to the scrambled control (37.2% colocalisation, Fig. 8A ix-xii) , resembling the state of hnRNP A1 in mockinfected cells (80.1% colocalisation, Fig. 8A i-iv) . These data demonstrated that the silencing of MINK reduced the hnRNP A1 signals in the cytoplasm as the degree of colocalisation between the hnRNP A1 signals and DAPI signals in the nucleus increased upon the siRNA knockdown of MINK.
Similarly, we investigated the subcellular localisation of hnRNP A1 upon the inhibition of p38 MAPK signalling pathway with the use of a p38 MAPK inhibitor (SB203580). Fig. 8B shows the immunofluorescence staining of hnRNP A1 in EV71-infected cells at 8h post-infection. The level of the hnRNP A1 in the nucleus increased significantly upon SB203580 treatment at 50μM (81.3% colocalisation, Fig. 8B ix-xii) compared to the DMSO control (52.9% colocalisation, Fig. 8B v-viii). In addition, the level of hnRNP A1 in the nucleus was even higher in cells treated with 100μM of SB203580 (92% colocalisation, Fig. 8B xiii-xvi), resembling the state of hnRNP A1 in mock-infected cells (94.5% colocalisation, Fig. 8B i-iv). These data demonstrated that the inhibition of p38 MAPK reduced the hnRNP A1 signals in the cytoplasm as the degree of colocalisation between the hnRNP A1 signals and DAPI signals in the nucleus increased upon SB203580 treatment. Together, we have shown in our study that it is likely that hnRNP A1 was one of the targets downstream of MINK/p38 MAPK signalling which is involved in promoting the translation efficiency of EV71 IRES as the relocalisation of hnRNP A1 to the cytoplasm where it binds to the IRES of EV71 RNA is required to facilitate translation initiation of viral RNA.
doi:10.1371/journal.ppat.1004686.g007 (red), an IRES-transacting factor, was investigated by indirect immunofluorescence assay. Immunofluorescence detection of double-stranded RNA (dsRNA, green) with the nuclei stained with DAPI (blue) was shown to indicate EV71 infection. The images were taken at 100X magnification. Colocalisation quantification was based on the Manders Overlap Coefficient (MOC) using whole-cell immunofluorescence (WCIF) ImageJ software [36] and represented as percent colocalisation at the respective siRNA concentrations. Error bars represent the standard deviation of duplicate data. (B) RD cells were subjected to infection with EV71 and post-treated with SB203580 (p38 MAPK inhibitor) for 8h. SB203580-treated cells were fixed and the subcellular localisation of hnRNP A1 (red) was investigated by indirect immunofluorescence assay. Mock-infected and DMSO-treated cells were included as infection and solvent control, respectively. The images were taken at 100X magnification. Colocalisation quantification was based on the MOC using WCIF ImageJ software and represented as percent colocalisation at the respective drug concentrations. Error bars represent the standard deviation of duplicate data. A1 (S4H Fig). Thus, it is likely that the cytoplasmic localisation of hnRNP A1 as a result of MINK/p38 MAPK signalling was not triggered by Mnk. Further investigation is required to identify the p38 substrates involved downstream of MINK/p38 MAPK signalling that modulates the cytoplasmic localisation of hnRNP A1.

Discussion

Infected cells have been shown to induce the activation of kinases in early events of viral infection, such as the activation of PAK1 upon Myxoma Virus infection [44] . Since MINK is a MAP4K, an upstream regulator in the MAPK signalling cascade, it was hypothesised that MINK might have been activated by early events of viral infection, such as the attachment of the virus or the uncoating of the virions. Activation profiles of MINK and entry assays have revealed that virus attachment and clathrin-mediated entry was not sufficient in triggering the activation of MINK. Instead, active replication of the virus and the accumulation of viral RNA or proteins might be potential inducers of MINK phosphorylation. To determine the exact cellular or viral factor that activated MINK upon infection, further studies such as a co-immunoprecipitation could be employed to identify viral factors that interact with MINK protein to trigger its activation.
Investigation into the stage of involvement of MINK in EV71 replication has also revealed interesting findings on the functional role of MINK in the propagation of EV71. Prior to our work, MINK has been suggested to play a role in the reorganisation of cytoskeleton such as actin filaments [22] which have been implicated in the formation of clathrin-coated vesicles in endocytosis pathways [45, 46] . Since EV71 is known to enter RD cells via clathrin-mediated endocytosis [26] , we first hypothesised that actin rearrangement induced by the activation of MINK may play a role during the clathrin-mediated entry of EV71. However, results from the viral entry study via viral RNA transfection were contrary to this hypothesis as silencing of MINK continued to cause inhibition of progeny virus production despite bypassing the normal entry pathway. Results from the real time RT-PCR analysis have also indicated that the silencing of MINK has an effect on the viral RNA levels. During EV71 replication, protein translation of the viral genomic RNA has to precede viral RNA replication as the production of non-structural proteins such as the RNA-dependent RNA polymerase (Protein 3D) is essential for the synthesis of viral RNA. As such, synthesis of viral proteins was also investigated upon the silencing of MINK, structural viral protein (VP0 and VP2) and non-structural protein (3D) levels was significantly inhibited in a dose-dependent manner with increasing MINK siRNA concentrations. However, since VP0 and VP2 are both products of proteolytic cleavage of the viral polyprotein, there is a possibility that the siRNA knockdown of MINK affected the proteolytic cleavage efficiency which led to the reduction in VP0 and VP2 levels. The VP2 antibody (MAB979) used in this study recognizes a specific epitope on the VP2 protein [47] , hence, larger incomplete processed viral polyproteins containing the VP2 epitope can also be detected. In this study, we have demonstrated that the protein levels of larger incomplete processed viral polyproteins (sizes coincided to that of P1 and VP4+VP3+VP2) [48] also showed similar trend to that of VP0 and VP2 protein levels where a dose-dependent reduction in the incomplete processed viral polyproteins levels was observed with the siRNA knockdown of MINK but not in the scrambled control. This narrowed down the potential target event of MINK involvement in the process of viral protein synthesis and not the proteolytic cleavage of the viral polyprotein. In addition, siRNA knockdown of MINK did not lead to an accumulation of intracellular virus, but instead, resulted in a global decrease in both extracellular and intracellular virus titres. This demonstrated that the silencing of MINK suppressed the production of progeny virus and not virus release.
Although the functional relationship between MINK and p38 MAPK in normal cellular processes [23, 49] has already been established, this association has not been reported in virus replication. Here, activation profile of p38 MAPK upon EV71 infection has suggested its essential role in EV71 replication and siRNA-mediated gene knockdown has demonstrated the correlation between MINK protein expression (consequently its activation) and p38 MAPK activation in the context of virus replication. The activation profile of MINK and p38 MAPK has also ascertained that p38 MAPK was activated downstream of MINK, which corresponded to what was reported in uninfected cellular conditions [23, 49] .
Involvement of p38 MAPK signalling pathway in virus replication has been well established and this signalling cascade has been reported to promote viral RNA synthesis and protein synthesis in some viruses such as the encephalomyocarditis virus (EMCV) [25] , mouse hepatitis virus (MHV) [50] and hepatitis B virus [51] . As shown previously by Hirasawa and his colleagues (2003), p38 MAPK signalling pathway promotes viral protein synthesis but not viral RNA synthesis in EMCV, which also belongs to the same family as EV71. Their study has demonstrated that p38 MAPK signalling pathway facilitates EMCV protein synthesis by promoting the translation efficiency of the IRES in EMCV. Supporting this hypothesis, our study has demonstrated the crucial role of p38 MAPK in the propagation of EV71 and that the activation of the MINK/p38 MAPK signalling pathway promotes the translation efficiency of EV71 IRES during EV71 replication. Time-of-addition assay conducted in our study verified that p38 MAPK inhibitor, SB203580, was effective in inhibiting the early events of EV71 replication cycle post-infection and not late events such as virus release. Inhibition of p38 MAPK signalling also demonstrated an inhibition in the synthesis of EV71 viral proteins (3D, VP0 and VP2) and viral RNA replication which were in agreement with the results observed upon the silencing of MINK. Contrary to what was observed with coxsackievirus B3 (CVB3) [31] , the inhibition of p38 MAPK did not block the release of the virus. Hence, this further supported the relationship between MINK and p38 MAPK signalling and the involvement of MINK/p38 MAPK in EV71 protein synthesis which resulted in a global decrease in the progeny virus production.
As mentioned in this study, EV71 has a type 1 IRES that requires eukaryotic initiation factors (eIFs) and IRES-specific transacting factors (ITAFs) to initiate viral protein translation [52] . Previous study on a poliovirus/rhinovirus chimera (PSRIPO) [34] has shown that signal transduction to Mnk1, a downstream substrate of p38 MAPK can favour viral, cap-independent translation via eIF4E phosphorylation and expression. Contrary to this finding, our data showed a down-regulation in the protein expression of eIF4E which led to a consequent reduction in its phosphorylation levels, suggesting that eIF4E phosphorylation and expression may not be crucial for EV71 replication. This finding was supported by another study [53] that has demonstrated that induction of miRNA-141 during EV71 infection down-regulated eIF4E in an attempt to suppress cap-dependent translation and promote the switch to cap-independent translation.
Members of the heterogeneous nuclear ribonucleoprotein (hnRNP) classes have been identified as trans-acting factors that control translation initiation of various cellular and viral mRNAs at the IRES [60] . Among the hnRNP family, hnRNP A1 has been reported to modulate the IRES-mediated viral protein translation of various viruses such as the human rhinovirus (HRV) [61] and EV71 [20] . Although, hnRNP A1 localises predominantly in the nucleus, it is able to shuttle between the nucleus and cytoplasm in a regulated manner [62] . Infection of cells with HRV and EV71 has shown to result in the cytoplasmic relocalisation of hnRNP A1 where it interacts directly with the viral IRES sequences [20] . Apart from picornaviruses, cytoplasmic accumulation of hnRNP A1 has also been reported to play a role in the positive regulation of human immunodeficiency virus (HIV) [63] and Sindbis virus (SINV) [20] viral RNA translation. In uninfected cells, activation of the p38 MAPK pathway upon osmotic shock or UV irradiation has been revealed to result in a phosphorylation-dependent cytoplasmic accumulation of hnRNP A1 [35] . Furthermore, a separate study has also demonstrated that the p38 MAPK interacts and regulates the subcellular localisation of hnRNP A1 in a Mnk1-dependent manner in senescent cells [37] . The cytoplasmic relocalisation of hnRNP A1 after EV71 infection may therefore also be dependent on the p38 MAPK pathway and its downstream substrate Mnk1 as in uninfected cells. In our study, subcellular localisation studies unravelled the relationship between MINK protein expression and hnRNP A1 localisation in the cells. Interestingly, we have found that the silencing of MINK upon EV71 infection did not result in the cytoplasmic accummulation of hnRNP A1 which was usually observed in infected cells. The nuclear retention of hnRNP A1 could be due to either lower levels of EV71 replication as a result of MINK silencing or a block in nuclear export signal [64] brought about by the siRNA knockdown of MINK. Similarly, the inhibition of p38 MAPK with a specific p38 MAPK inhibitor (SB203580) also resulted in the accumulation of hnRNP A1 signals in the nucleus. Although we have demonstrated that inhibition of the MINK/p38 MAPK signalling pathway reduced the hnRNP A1 signals in the cytoplasm that was observed in control EV71-infected cells, we have no direct evidence suggesting that MINK plays a direct role on the cytoplasmic relocalisation of hnRNP A1 where it binds directly to the IRES sequences of the viral genome to promote the IRES-mediated translation of the EV71 viral RNA. In addition, we have also shown that despite its activation during EV71, p38 MAPK substrate Mnk1 was not involved in the regulation of EV71 protein synthesis and the cytoplasmic relocalisation of hnRNP A1. Hence, the exact mechanism of how MINK/p38 MAPK signalling pathway affected the cytoplasmic relocalisation of hnRNP A1 during EV71 infection and the p38 MAPK substrates involved needs to be further established. Nonetheless, we have shown in this study that a novel host kinase (MINK) mediates the cap-independent translation of EV71 RNA, possibly by modulating the subcellular localisation of hnRNP A1, which further supports its propagation (Fig. 9 ). As such, MINK can be further explored as potential antiviral target for the inhibition of EV71 viral replication at the viral protein translation stage.

siRNA reverse transfection

A reverse transfection protocol was used to perform the siRNA screen. The primary screen was performed in a 384-well format at a final concentration of 25nM siRNA using lμl of Dharma-FECT-1 transfection reagent (Thermo Scientific) per well. Specific targeting siRNAs and scrambled siRNAs were dissolved in diethyl pyrocarbonate (DEPC)-treated reverse osmosis (RO) water to a stock concentration of 100μM. The siRNAs were then diluted to desired working concentrations with DharmaFect Cell Culture Reagent (DCCR) and DharmaFect-1 transfection reagent. The siRNAs were transfected into RD cells for 72h at 37°C in a 5% CO 2 atmosphere prior to infection.

Immunofluorescence assay

For the study on hnRNP A1 localisation, RD cells grown on coverslips were pre-treated with either scrambled or MINK siRNA for 72h. At 72h post-transfection, the cells were infected at MOI 1 for 8h. After washing thrice with PBS, the mock-infected and virus-infected cells on the coverslip were fixed with 4% paraformaldehyde (Sigma-Aldrich) at room temperature for 15 min. After three washes with PBS, the cells on the coverslip were permeabilised in 0.1% Triton X-100 at room temperature for 10 min and washed for another three times with PBS. The samples were then blocked in PBS containing 5% bovine serum albumin (BSA, MP Biomedicals) for 1h at 4°C and then incubated with primary antibodies at appropriate dilution: rabbit monoclonal anti-hnRNP A1 antibody (#ab177152, Abcam) and mouse monoclonal anti-dsRNA antibody (SCICONS) at 37°C for 1h. Upon removal of the primary antibodies, the samples were washed thrice with PBS. The samples were then incubated with secondary antibodies at appropriate dilution: Rhodamine-conjugated goat anti-rabbit IgG (Millipore), fluorescein isothiocyanate (FITC)-conjugated goat anti-mouse IgG (Millipore) at 37°C for 1h. Subsequently, Duolink In Situ mounting medium with DAPI (OLink BioSciences) was used to stain the cell nuclei and mount the coverslip on a microscope slide. The specimens were viewed with a 100X oil immersion lens with a numerical aperture (NA) of 1.6 of Olympus IX81. Colocalisation was quantified based on fluorescence microscopy images using the NIH ImageJ software (Wright Cell Imaging Facility) via the colocalisation analysis plug-in. Manders overlap coefficient (MOC) represents the proportion of normalised pixels in which the two signals overlap and was used as a measure of colocalisation [36] . MOC ranges from 0 for no colocalisation between the signals and 1 for perfect overlap. The percentage of colocalisation between the hnRNP A1 signals and nucleus was determined based on the MOC.

Viral plaque assays

Plaque assay was performed on monolayers of RD cells in 24-well plates for the quantification of virus titre. Supernatants from EV71-infected samples were diluted in 10-fold dilutions with DMEM supplemented with 2% FCS before infection. The cells were incubated with the supernatant for 1h at 37°C with gentle rocking during the adsorption period. Infected cells were washed twice with PBS and overlaid with 1% carboxymethyl-cellulose (CMC) in DMEM with 2% FCS. Plaques were allowed to form for 4 days at 37°C in an atmosphere of 5% CO 2 . After which, the cells were fixed and stained with 10% paraformaldehyde/1% crystal violet (Sigma-Aldrich) solution. Virus titres were expressed as plaque forming units per millilitre (PFU/ml).

Cell viability assay

Cell viability profiles of siRNA-treated or drug-treated cells were assessed using the AlamarBlue reagent (Invitrogen) as recommended by the manufacturer's protocol. AlamarBlue reagent was added to each well and the plates were incubated at 37°C supplemented with 5% CO 2 for 4h and fluorescence measurements were taken using the Infinite 200 series microplate reader (Tecan). The measurements were performed at excitation wavelength of 570 nm and emission wavelength of 585 nm.

Viral RNA transfection into RD cells

Viral RNA was isolated and purified from EV71 viral supernatants using the QIAamp viral RNA minikit (Qiagen) according to the manufacturer's instructions. siRNA-treated RD cells were transfected with 2μg of EV71 viral RNA. During the transfection, EV71 viral RNA was diluted in reduced serum OPTI-MEM I (Gibco, Invitrogen) with the addition of Plus reagent (Invitrogen) added and left to incubate at room temperature for 5 mins. The diluted viral RNA solution was then added to the diluted Lipofectamine LTX and incubated at room temperature for 30 min. Growth medium was removed from the wells on the 24-well plate and 1ml of DMEM supplemented with 2% FCS was added into each well. Following incubation, viral RNA-Lipofectamine mixture was added into each well, giving a final amount of 2μg of viral RNA per well. Supernatants and cell lysates were harvested at indicated time-points post-transfection.

Quantitative reverse transcription-PCR (qRT-PCR)

Total cell lysate was harvested at 0h, 8h and 10h post-infection and extraction was carried out using Total RNA Mini Kit (Blood/ Cultured Cell) (Qiagen). Extracted RNA was then subjected to Reverse-Transcription Real-Time Polymerase Chain Reaction (qRT-PCR).

Treatment of cells with p38 MAPK inhibitor (SB203580) and Mnk1 inhibitor (CGP57380)

A working concentration of 10mM SB203580 (#5633S, Cell Signalling Technology) and CGP57380 (ab120365, Abcam) was prepared in dimethyl sulfoxide (DMSO). RD cells were infected with EV71 at MOI 1 for 1h before drug treatment for another 12h. After 12h of incubation, the supernatant were harvested for plaque assay.

Time-of-addition studies

Time-of-addition assay was performed for SB203580 (#5633S, Cell Signalling Technology) on EV71-infected RD cells in 96-well plates. Cells treated with 0.5% DMSO were used as control. For the pre-treatment assay, cell monolayers were treated with 50μM of SB203580 for 2h at 37°C before being washed twice with PBS and infected with EV71 at MOI 1. After the 1h virus adsorption period, infected cells were washed with PBS and incubated in DMEM supplemented with 2% FCS at 37°C with 5% CO 2 for 12h before supernatants were harvested for viral plaque assay.
For the co-treatment assay, SB203580 was added together with EV71 at MOI 1 to obtain a final SB203580 concentration of 50μM. After incubating the cells with this mixture for 1h, infected cells were washed with PBS and incubated in DMEM supplemented with 2% FCS at 37°C with 5% CO 2 for 12h before supernatants were harvested for viral plaque assay.

Statistical analysis

The Z' factor, a statistical measurement of the distance between the standard deviations for the signal versus the noise of an assay, was employed as an indicator for the robustness of the screen. Experiments to determine the Z' factor was conducted in a 384-well plate using positive controls where virus-infected cells were not treated with siRNA (growth media, transfection reagent and DCCR) and mock-infected cells as negative controls. The Z' factor was then computed using the equation: 1-(3 x S.D. positive control + 3 x S.D. negative control) / (mean positive control-mean negative control). In other studies, one-way ANOVA test was used to compare the data and the results were considered to be significant if p 0.05.

Accession numbers

Supporting Information S1 Table. List of targeted genes in the human serine/threonine kinase siRNA library used in primary screen. Band intensities of 3D protein upon siRNA knockdown of MINK. The band intensities representing 3D protein expression level were quantitated with reference to actin control bands (for each siRNA concentration) and 0nM using ImageJ Gel Analysis program. (C) Band intensities of 3CD protein upon siRNA knockdown of MINK. The band intensities representing 3CD protein expression level were quantitated with reference to actin control bands (for each siRNA concentration) and 0nM using ImageJ Gel Analysis program. (D) EV71 3D protein expression levels upon SB203580 treatment. EV71-infected RD cells were treated with SB203580 and cell lysates were harvested for Western blotting at 8h post-treatment. 3CD and 3D viral protein expression was observed to decrease with increasing concentration of the p38 MAPK inhibitor. (E) Band intensities of 3D and 3CD upon SB203580 treatment. The band intensities representing 3D and 3CD expression level were quantitated with reference to actin control bands (for each concentration) and 1.0% DMSO control using ImageJ Gel Analysis program. β-actin was included as a loading control. (D) Quantification of phospho-eIF4E (S209) and total eIF4E protein bands with reference to actin control bands (for each CGP57380 concentration) and untreated control using ImageJ Gel Analysis program. (E) Viral protein expression levels upon CGP57380 treatment. EV71-infected RD cells were treated with CGP57380 and cell lysates were harvested for Western blotting at 8h post-treatment. Constant VP0 and VP2 viral protein expression was observed with increasing concentration of the Mnk1 inhibitor. (F) Band intensities of VP0 and VP2 upon CGP57380 treatment. The band intensities representing VP0 and VP2 protein expression level were quantitated with reference to actin control bands (for each concentration) and 1.0% DMSO control using ImageJ Gel Analysis program. (G) Cell viability of CGP57380-treated cells and untreated control cells were measured using alamar-Blue assay at 12h post-treatment. Values obtained were normalised against 1.0% DMSO control. Virus titres in the supernatant of cells (denoted by bars) treated with varying concentrations of CGP57380 post-adsorption were analysed via viral plaque assay. Error bars represent standard deviation (SD) of triplicate data. Statistical analyses were performed using one-way ANOVA and Dunnett's test (Graphpad software) against untreated control P > 0.05 (n = 3) versus 1.0% DMSO control (H) RD cells were subjected to infection with EV71 and post-treated with CGP57380 (Mnk1 inhibitor) for 8h. CGP57380-treated cells were fixed and the subcellular localisation of hnRNP A1 (red) was investigated by indirect immunofluorescence assay. Immunofluorescence detection of double-stranded RNA (dsRNA, green) with the nuclei stained with DAPI (blue) was shown to indicate EV71 infection. The images were taken at 100X magnification. Colocalisation quantification was based on the MOC using WCIF Ima-geJ software and represented as percent colocalisation at the respective drug concentrations. Error bars represent the standard deviation of duplicate data. (PPTX)

Replication of EV71 triggers MINK phosphorylation

To further ascertain that initial binding of the virus was not required for the activation of MINK, an infection inhibition assay was performed. It has been well-established that scavenger receptor class B2 (SCARB2) is the cellular receptor for EV71 on RD cells [28] , hence an antibody to SCARB2 was used to block the cellular SCARB2 receptors to prevent virus binding. Viral plaque assay was first conducted to confirm the efficacy of the anti-SCARB2 antibody in blocking EV71 infection. A corresponding control IgG antibody that does not bind specifically to any proteins was used as a negative control. Results from the infection inhibition assay showed that pre-treatment of RD cells with the anti-SCARB2 antibody blocked EV71 infection in a dose-dependent manner as the virus titre was significantly reduced by~1.5 log at the highest concentration of anti-SCARB2 antibody used (Fig. 4C ). On the other hand, treatment with the control IgG antibody did not affect the binding and entry of virus into the cells (Fig. 4C ) as the virus titres remained constant across increasing concentrations of IgG antibody.

Silencing of MINK and p38 MAPK inhibition reduced hnRNP A1 signals in the cytoplasm upon EV71 infection

In a recent publication, phosphorylated eIF4E has been reported to improve viral protein translation via the IRES of rhinovirus by modulating the eIF4G:IRES interaction [34] . Hence, to investigate if eIF4E was the downstream effector of the MINK/p38 MAPK pathway that led to the increase in IRES-mediated protein translation efficiency, a time course study was conducted to examine the activation profile of downstream substrates of p38 MAPK in response to EV71 infection. Contrary to the findings in the rhinovirus IRES-mediated protein translation, we did not detect any increase in eIF4E phosphorylation levels upon EV71 infection. Instead, eIF4E protein expression and phosphorylation levels were observed to decrease upon EV71 infection (S4A and S4B Fig), suggesting that eIF4E might not be essential for EV71 replication.
Cytoplasmic localisation of hnRNP A1 resulted from MINK/p38 MAPK signalling was not stimulated by Mnk1 activity As Mnk1 has been implicated in the cytoplasmic localisation of hnRNP A1 downstream of p38 MAPK signalling in uninfected cells [37] , we were interested to know if Mnk1 was the p38 MAPK substrate that triggered the increase of hnRNP A1 signals in the cytoplasm during EV71 infection. A time course study was first conducted to examine the activation profile of Mnk1. Compared to the mock-infected cells, EV71 infection resulted in increased Mnk1 phosphorylation levels at 8h post-infection with relatively constant Mnk1 protein expression levels (S4A and S4B Fig). To determine if Mnk1 phosphorylation was triggered by the replication of the virus, RD cells were exposed to UV-inactivated EV71. Cells exposed to UV-inactivated EV71 showed constant phosphorylation level of Mnk1 throughout the 12h time course (S4A and S4B Fig) . This indicated that attachment of the virions to cell surface receptors or virus entry process was not sufficient to trigger the phosphorylation of Mnk1, which corresponded to what was observed previously for the activation of MINK and p38 MAPK upon EV71 infection ( Fig. 4 and 5) .

Discussion

Viruses have been known to hijack cellular signalling pathways during viral infection to facilitate events such as viral entry, inhibition of apoptosis and to escape antiviral activities elicited by host factors such as interferon [38, 39, 40] . In relation to EV71, the recent work conducted by Hussain and colleagues (2011) identified several kinases with potential involvement in the clathrin-mediated endocytosis of EV71 into RD cells [26] . Thus, host kinases may represent a family of host factors which may play roles in facilitating EV71 replication and have the potential of becoming a viable antiviral target.
In this study, a novel mitogen-activated protein kinase kinase kinase kinase (MAP4K) has been identified from our primary screen for further investigation as interest in the MAPK family of kinases was sparked by previous studies on the involvement of MAPK family members in viral replication [41] and the pathogenesis of viruses [42, 43] . Here, we demonstrated that MINK plays an essential role in EV71 replication and may be a potential target for antiviral development.

Cell line and viruses

Human rhabdomyosarcoma (RD) cells (CCL136TM, ATCC) cells were maintained in Dulbecco's Modified Eagle's Medium (DMEM, Sigma-Aldrich) enriched with 10% fetal calf serum (FCS, PAA Laboratories) in T75 at 37°C in an atmosphere of 5% CO 2 . Human Enterovirus 71 (EV71) strain H (VR-1432) was obtained from ATCC (GenBank accession no. AY053402.1) and EV71 strain 5865/sin/000009 (designated as strain 41, GenBank accession no. AF316321) was a kind gift from Dr Vincent Chow [65] , Department of Microbiology, National University of Singapore. Coxsackivirus A6 (CA6, GenBank accession no. KC866983) and Echovirus 7 Wallace strain (Eo7-Wallace, GenBank accession no. AF465516) were obtained from the department collection and the viruses were propagated in RD cells. UV-inactivated EV71 was prepared by subjecting the virus stock to UV light irradiation for 2h before performing viral plaque assay to ensure complete inactivation.

Transfection of bicistronic construct of EV71-IRES

The bicistronic construct of EV71 IRES (Fig. 7A ) was a kind gift from Professor Peter C McMinn, University of Sydney. The construct contains the 5' untranslated region (UTR) of the EV71-26M strain and two reporter genes, Renilla luciferase (RLuc) and firefly luciferase (FLuc). The RLuc-reporter gene was positioned upstream of the EV71 5' UTR controlled by the cytomegalovirus promoter (CMV). The firefly luciferase (FLuc) reporter gene is ligated downstream of the 5' UTR which controls its expression [67] .
45 section matches

Abstract

Viruses are fast evolving pathogens that continuously adapt to the highly variable environments they live and reproduce in. Strategies devoted to inhibit virus replication and to control their spread among hosts need to cope with these extremely heterogeneous populations and with their potential to avoid medical interventions. Computational techniques such as phylogenetic methods have broadened our picture of viral evolution both in time and space, and mathematical modeling has contributed substantially to our progress in unraveling the dynamics of virus replication, fitness, and virulence. Integration of multiple computational and mathematical approaches with experimental data can help to predict the behavior of viral pathogens and to anticipate their escape dynamics. This piece of information plays a critical role in some aspects of vaccine development, such as viral strain selection for vaccinations or rational attenuation of viruses. Here we review several aspects of viral evolution that can be addressed quantitatively, and we discuss computational methods that have the potential to improve vaccine design.

Vaccine development strategies

Live attenuated vaccines (LAVs). Immunization with LAV has been proved to be the most efficient vaccination strategy to date [34, 35] . LAV preparations include viruses with reduced virulence, which means that they do not produce the disease when infecting the host, or they produce a mild version of the disease. Viruses become attenuated for the original host after serial infections (passages) in cell culture of different organism. This is the case of the polio vaccine preparation in monkey cells [36] . Infection of embryonated hen eggs is the standard protocol to obtain attenuated yellow fever or measles viruses suitable for vaccine preparation [37] . The rationale for this attenuation strategy is that due to the high error rates during replication of viruses, especially of RNA viruses [2] , the virus accumulates mutations in the genome that optimize their replication in the new host or new cell type, at the expense of replication efficiency in the original host [1, 34, 38] .

Phylogenetic methods in influenza vaccine design

With one exception, the few attempts to bring HIV vaccines to the last phases of clinical trials have been quite disappointing so far. One of the first and most prominent vaccine candidates, the VAXGEN vaccine, was intended to immunize subjects with a recombinant envelope protein of HIV (rgp120) [93, 94] . The envelope protein is located on the surface of HIV and is responsible for the attachment of the virus to the host cell surface receptor [95] . Antibodies targeted against this protein could block HIV infection and subsequently block virus entry into cells. During VAXGEN trials, the immunization induced the production of antibodies in vaccinated individuals, but they were unable to control infection or viremia.
The STEP vaccine trials were designed to test the efficiency of a T cell vaccine in reducing viremia and enhancing the cellular immune response [96] . This vaccine candidate is a therapeutic vaccine (see below), because it was intended to enhance the immune response against HIV even if sterilizing immunity would not be achieved. The STEP vaccine formulation was based in an adenovirus serotype 5 vector (Ad5). The vaccine included three independent Ad5 vectors, each one carrying one of the three HIV proteins Gag, Pol, and Nef. Although in early phases of the clinical trials, the vaccine was proven to elicit specific anti-HIV T cell responses, no significant protection was conferred to people receiving the vaccine in phase IIb trials. Indeed, individuals having immune memory against the Ad5 vector were more susceptible to infection by HIV [97] . This unexpected result was probably due to an adding-fuel-to-the-fire effect, in which the Ad5 vector activated T cells creating a suitable environment for HIV replication [98] .

Attenuation strategies and the evolution of virulence in RNA viruses

Since attenuated viruses may elicit a potent immune response without causing harm to the host, several strategies are being explored to obtain candidate viruses for LAVs, i.e., viruses displaying a reduced cell killing or replicative ability. Serial cytolytic transfers in cell culture tend to select viruses attenuated in the original host [160] . This is the case for FMDV or yellow fever virus among other viruses, and is currently an approved technique for several LAV preparations. Viruses selected after severe bottlenecks, such as serial plaque-to-plaque transfers, present a reduced fitness due to the accumulation of mutations associated with the Müller's ratchet effect [161, 162] . A new promising strategy for the attenuation of viruses is the rational design and synthesis of viral genomes with a strong codon bias. This approach has been implemented for Poliovirus and influenza virus [163, 164] . The viral genome synthesized encodes the same amino acid sequence as the wild type virus but encoded by infrequent codons in their host cells. Viruses harboring fidelity mutations in the replicase genes tend to produce a quasispecies of lower diversity and to be attenuated in vivo. This feature has also been employed to the rational design of a Poliovirus LAV [165] . Other strategies conceived to limit viral replication include the design of specific microRNAs or zinc finger nucleases targeting the viral genome [166] .

Conclusions

Viral evolution and the genetic diversity it produces are fundamental factors for the success of vaccine candidates, because immune responses need to be stimulated against a potentially very broad spectrum of existing viruses and new viral immune escape variants are likely to be generated. Mathematical modeling of viral evolutionary dynamics will therefore play an increasingly important role in vaccine design. It can identify genomic regions that are under selective pressure, support the selection or construction of vaccine strains, predict evolutionary escape from immune pressure, guide vaccination campaigns, estimate the effect of therapeutic vaccines, and support the design of new attenuation strategies. Most of our discussion has been in the context of RNA viruses and many issues are most pronounced for this class of viruses. Nevertheless, we expect most statistical models and computational methods to be applicable to other viruses and different pathogens, too. On the other hand, the distinct evolutionary dynamics of influenza A and HIV-1, two of the most widely studied RNA viruses, have highlighted the need for careful analysis of viral infection dynamics within and among individuals.

Viruses are intracellular parasites that need the cellular machinery of the host to reproduce [1] . They have the potential to generate huge population sizes in short generation times. Viruses in general, and RNA viruses in particular, exist in genetically heterogeneous populations because of their error-prone replication [2] . These features make the evolution of viruses a phenomenon observable on short time scales of weeks to months. The consequences of the extreme viral evolutionary dynamics are of tremendous importance for disease control and prevention. For example, influenza vaccines need to be updated every year, viral variants develop resistance to antiviral drugs, and mild viral strains turn into virulent ones spontaneously. These global health care issues and others arise from the rapid evolution of viruses.

Phylogenetic methods in influenza vaccine design

The application of statistical genetics and phylogenetics methods to influenza sequence data has not only improved our understanding of the evolutionary dynamics and the epidemiology of the virus, but it has also become an integral part of the yearly vaccine design cycle. However, the successful case of influenza does not seem to provide a practical model for HIV. One reason for this discrepancy might be the evolutionary dynamics of HIV which are strikingly different from those of influenza. Rather than the drift-and-shift pattern of influenza evolution which generates only a small amount of genetic diversity around the successful trunk lineage, HIV tends to spread out from an ancestor in a radial fashion and to generate much more variation. The worldwide diversity of influenza sequences in any given year appears to be comparable to the diversity of HIV sequences found within a single infected individual at one time point [62] . Thus, an HIV vaccine must stimulate a very broad reactive immune response against a large set of diverse viral strains and the genetic makeup of these sequences is much more difficult to predict from the currently circulating strains as compared to influenza. It is for these and possibly other reasons that the same bioinformatics-assisted vaccine design approach that is established for influenza, has not been equally successful for HIV to date. In the following sections, we will discuss extensions of the models discussed above as well as complementary mathematical and computational approaches that might be of help in search for an HIV vaccine in the future.
Moreover, HIV has a great capacity for recombination among subtypes, and as a consequence, new circulating recombinant forms are constantly arising [89, 90] . Each subtype itself represents a great genetic diversity as well and even during infection of a single patient, HIV exists as an ensemble of different sequences [88] . The high genetic diversity of HIV populations hampers the development of a vaccine of broad applicability. Vaccine candidates need to elicit responses against multiple epitopes in order to counteract the immune evasion by virus mutation [91, 92] .

Molecular profiling techniques, including DNA sequencing, have produced an enormous amount of data on viral spread, genetic diversity, and infection dynamics. The integration and analysis of these data can provide valuable information on the evolution of viral pathogens. Mathematical, statistical, and computational methods are necessary to deal with those large data sets and to predict phenotypes form genetic data that ultimately can be used in vaccine development. In this review, we summarize some computational and mathematical techniques that play a critical role in understanding viral evolution and vaccine design. Specifically, we discuss phylogenetic methods for vaccine strain selection, statistical models of evolutionary escape from selective immune pressure, and virus dynamics models for therapeutic vaccines and attenuation strategies. Our major examples are Influenza, human immunodeficiency virus (HIV), and foot-and-mouth disease virus (FMDV). They are all RNA viruses of great medical or veterinary importance and have been studied extensively.

Viral evolution

A quantitative description of viral evolution is necessary for monitoring the spread of viral pandemics and for developing effective therapies and vaccines [3] . Viruses are not only a threat to human health, but they also provide attractive model systems for evolutionary studies due to their short genomes, large population sizes, and high genetic diversity [4] . The extreme replication dynamics of RNA viruses, for example, allow for observing significant evolutionary changes over time. Hypotheses and theories about evolutionary mechanisms can often be tested directly with these measurably evolving viruses [5] .
The mutation rate of RNA viruses is about a million times larger than the human mutation rate [6] . Thus, RNA viruses display a huge genetic diversity and this feature is critical for survival of the virus [7, 8] . Virus populations are exposed to fluctuating environments when migrating through different organs and tissues of the host organism and when exposed to immune responses mounted by the host. Transmission to a new host is typically associated with both traversing various tissues and facing new immune responses, and therefore it represents a major bottleneck for the virus population. The genetic diversity of RNA viruses makes it likely that adapted variants preexist in the population even before the selective pressure has changed [9, 10] .
Because the diversity of the virus population can determine its evolutionary fate, selection seems to operate on the population level rather than at the level of individual viruses [11] . This idea was originally developed for selfreplicating RNA molecules and termed quasispecies theory [12] , and then applied to RNA viruses [13, 14] . One prediction of quasispecies theory is the existence of an upper bound on the mutation rate beyond which the population cannot maintain its essential genetic information. Many RNA viruses appear to have mutation rates close to this error threshold [9, 15, 16] .
In general, the study of correlated evolution among genotypic and phenotypic traits or between traits and the environment across species is known as the comparative method [23] . It has been extremely successful in analyzing DNA sequence data and it is the basis for predicting phenotypes, such as protein structure or function, from genotypes. In the case of RNA viruses, which display a large genetic diversity between hosts, the different viral quasispecies take the role of the different species in the traditional application of the comparative method. Viral phenotypes of interest include immunological escape and drug resistance.

Vaccine development strategies

Inactivated vaccines (IVs). Viral stocks are susceptible to inactivation by some chemical and physical treatments. IVs consist of a concentrated viral stock that has been treated with a chemical reagent, such as binary ethylenimine or formaldehyde, which completely abolish virus replication [40] . Some viruses are difficult to attenuate because they may change their antigenic properties or they remain virulent after few passages in cell culture [41] . The latter is the case for FMDV, an animal virus from the picornavirus family. When a virus cannot be attenuated with sufficient reliability, inactivation has been proven to be a successful vaccination strategy, as documented for FMDV, influenza, or hepatitis A vaccines [41] [42] [43] [44] . Some security issues may arise however, if the chemical compound does not reach all virus particles, for example because some viral particles tend to form compact aggregates, and the preparation still contains a portion of live viruses. Inactivated viruses are usually not efficiently presented by MHC class I molecules, which stimulate the cellular immune response. IV preparations include adjuvants which are chemical compounds that act on antigen-presenting cells enhancing the immunogenicity of the vaccine [45, 46] . However, the strength and duration of the protection induced by IVs is usually lower than that obtained with LAVs.

Phylogenetic methods in influenza vaccine design

Influenza A virus is a negative-stranded RNA virus that infects about one fifth of the worldwide human population each year [56] . The viral genome consists of eight segments and is categorized by the serology and genetics of its two surface glycoproteins neuraminidase (NA) and hemagglutinin (HA). Several subtypes of both genes have been isolated from mammalian and avian hosts, including the two most recent pandemic strains H3N2 and H1N1 currently circulating in the human population and responsible for the 1968-1969 Hong Kong Flu and the 2009 Swine Flu, respectively.
Evolutionary escape of HIV from selective immune pressure HIV populations display a high genetic diversity due to the quasispecies nature of RNA virus replication [2, 87, 88] . HIV occurs in three main groups, the principal of which, group M, is composed of nine subtypes.
Vaccine design strategies based on whole viral protein sequences make extensive use of phylogenetics. In addition to the basic methods discussed above, HIV-specific probabilistic models of protein evolution have been constructed which allow for improved phylogenetic inference using likelihood or Bayesian methods [103] . The reconstructed phylogenetic tree can guide the selection of viral genomes to be included in the vaccine. Different selection strategies have been proposed to stimulate a broad immune response and to minimize the amount of sequence divergence between the antigen and contemporaneously circulating viruses. Natural strains that represent the total observed sequence space or derived consensus sequences have been selected as vaccine strains [104] [105] [106] . Probabilistic phylogenetic models also allow for inferring the DNA sequences at internal nodes of the tree which represent extinct common ancestors. The most recent common ancestor (MRCA) of a given set of viruses is the root of the phylogenetic tree for these sequences [107] . It has been proposed as a vaccine strain stimulating cross-reactive immune responses against all of its descendants [104, 108] . However, in asymmetric phylogenies, both the consensus sequence and the MRCA can perform poorly at minimizing the distance to contemporary strains [109] . To address this limitation, the center of tree node has been proposed. It is calculated as the node minimizing the least squares distance to all leaves of the phylogenetic tree [110] .
The usefulness of phylogenetic trees is limited in the presence of reticulate evolutionary events, such as hybridization, horizontal gene transfer, or recombination, which cannot be represented by a tree. For this situation, phylogenetic network models have been developed [111] . They generalize phylogenetic tree models and include reticulate networks and split networks [112] . In most RNA viruses, homologous recombination can occur when a cell is coinfected with two different strains. In HIV, multiple infections are common [113] and the recombination rate is on the order of 2 to 3 times per genome per replication cycle [114] . Several epidemiological circulating recombinant forms provide evidence for recombination in HIV. Intra-host evolutionary dynamics are also shaped by recombination affecting the generation of multidrug-resistant strains in treated patients [115, 116] and the development of immune escape variants. Efficient parsimony algorithms have been developed for computing recombination networks [117, 118] .

Attenuation strategies and the evolution of virulence in RNA viruses

Many mathematical models have been developed to describe the evolution of virulence in diverse viral populations. One conclusion of these theoretical studies is that virulence can increase in the population under a variety of conditions [167] . The basic model of virus dynamics, however, states that less cytopathic variants are more productive in the long term of the infection, because the abundance of both viruses and infected cells is inversely correlated with the cytopathogenicity, a, of the viral variants [144] . For R 0 > 1, Recently, a FMDV population has been reported to diversify into two genetically distinct subpopulations that also differ in virulence. The viral variants have been characterized phenotypically in considerable detail and their coevolutionary dynamics, when competing for the same cell pool in vitro, have been analyzed. The competition experiment has been described by an extension of the basic model of virus dynamics introduced above.

Vaccine development strategies

Vaccine development can be considered one of the biggest achievements of modern medicine. While some bacterial families share common biochemical features, such as surface lipopolysacharide and therefore can be treated with common drugs or antibiotics [24] , viruses are typically different enough between families or even inside families, making the development of broadly applicable antiviral drugs challenging [2, 25] . For example, the base analogue ribavirin alone or in combination with interferon, has been proven to act as an antiviral compound of broad activity in the clinical treatment of infections with hepatitis C virus (HCV), hepatitis B virus (HBV), and respiratory syncytial virus (RSV) [26] [27] [28] [29] [30] [31] .
Overall, only few antiviral drugs targeting viral proteins are currently approved for use in humans [32] . Viruses that can be fought with chemical compounds include HIV, influenza virus, RSV, HCV, HBV, and the herpes viruses, including herpes simplex virus, varicella-zoster virus (VZV), and cytomegalovirus.
However, currently the vast majority of viruses cannot be controlled today with any approved compound and frequently not even all subgroups inside the same viral species are drug sensitive. For example, the 2009 seasonal influenza A virus (H1N1) presents a natural resistance to oseltamivir [33] . The efficiency of antiviral compounds is usually hampered by the generation of drug resistant viruses. Moreover, some compounds may help control infection, but only rarely can the virus be completely cleared from the host organism. By contrast, vaccines can boost the immune system response and, in principle, achieve complete clearance of a virus from an infected host. Therefore, vaccines are considered the best weapon to fight viruses.
The attenuated virus is still competent for replication and it retains the ability to infect host cells. For this reason, LAVs can elicit different effector mechanisms of the immune system. The intracellular replication of the virus can stimulate cytotoxic CD8 + cells because they can be presented by major histocompatibility complex (MHC) class I molecules. Particles released outside the cell can also be presented by class II MHC molecules [39] . After immunization with the LAV, the immune system is exposed to multiple antigens of the virus in its native conformation. Once the infection with the LAV is cleared from the organism, the virus-specific immune cells remain as memory cells in the host. A future challenge with the wild type virus will trigger the correct response, i.e., predominantly cellular response or predominantly humoral response. LAVs are considered the most successful vaccines because the efficient and multiple stimulation of the immune system typically induces a potent and durable response [38] .
DNA vaccination is a strategy based on injecting a DNA construct directly into the host [54, 55] . Such DNA constructs, which code for the immunizing protein or other parts of interest of the virus, can be transcribed and translated into the cell. Therefore, expressed gene products can elicit an immune response by presentation of peptides by MHC class I and II molecules.

Phylogenetic methods in influenza vaccine design

Not only selection, but also neutral genetic drift seems to play an important role in the evolution of influenza virus. Both evolutionary forces, termed antigenic drift and antigenic shift, have been observed in human hosts over the last century. Antigenic drift refers to neutral evolutionary changes accumulating over time, whereas antigenic shift involves a change in genetic and serological properties of the virus due to new HA or NA subtypes.
The influenza vaccine needs to be redesigned regularly to account for genetic changes in the virus population. Normally, the changes are made in response to the antigenic drift of the virus. For example, between 1968 and 2001 the H3N2 component of the influenza vaccine was changed a total of 17 times [62] . The selection of viral strains to be included in the vaccine for the coming season is based on the antigenic properties of recent isolates, on epidemiological data, and on post-vaccination serological studies in humans.
HIV is a member of the retrovirus family with the ability to integrate its genome into the host cell genome [95] . Genome integration is another challenge to develop an effective vaccine because latently infected cells cannot be recognized by the immune system until the integrated provirus is activated [136] .

Virus dynamics and therapeutic vaccines

A viral infection is a complex molecular process, but it can be approximated by a few major steps ( Figure 2) . Initially, the virus needs to find a susceptible cell and enter into it. Once inside the cell, virus replication starts and viral offspring is released. Infected cells will eventually die. Released virus particles are either inactivated or they hit a new susceptible cell, in which case the infection cycle is closed and a chain reaction of sequential infections is triggered [144] [145] [146] . The dynamics of the viral replication cycle can be expressed in mathematical terms as follows:
This ordinary differential equation (ODE) system describes uninfected cells, x, being infected with efficiency b, infected cells, y, dying and releasing viral offspring at rate a, and free virus, v, being produced at rate k and inactivated at rate u. In the absence of viral infection, cells reproduce at rate l and die at rate d. Oversimplifying the role of the immune system, the immune cells, z, grow and die with rates c and b, respectively. They remove infected cells from the system with efficiency p. Each specific viral family may give rise to modifications of this model due to variations in its life cycle. But the ODE system is the core of a family of mathematical models that describe the turnover of viruses and cells during an infection.
Virus dynamics models have been successfully employed to the study of simian immunodeficiency viruses (SIV), HIV, and HBV, among others [144, [146] [147] [148] [149] [150] . The dynamics of this model are shown in [151] , in which infection is treated as a microepidemic and host cells play the role of infected or susceptible individuals. Whether the virus infection can spread in the cell pool or not depends on a condition very similar to the spread of an epidemic in a population of individuals. The parameters of the model must satisfy the inequality R 0 > 1, where R 0 is the basic reproductive number, defined as the number of newly infected cells that arise from any one infected cell when almost all cells are uninfected [144, 151] . For a' = cp/b, this number is given by For generic parameters, if R 0 > 1 uninfected cells become infected and produce progeny viruses exponentially. Activation of the immune system (a' > 1) reduces the value of R 0 and slows down the spread of the infection. At the beginning of the infection, before the immune response is mounted (a' ≈ 0), and after the initial peak of viral load, viruses and infected and uninfected cells reach a stable equilibrium termed viral set point ( Figure 3 ). While monitoring viral load of SIV in infected macaques, a correlation between the viral load at initial stages of the infection and the viral set point was observed [152, 153] . One can demonstrate that the equilibrium abundance of viruses, v*, and the logarithm of the virus load during the exponential growth phase, follow the linear relationship This result is important in HIV research, because several studies indicate that there is a positive correlation between viral load and disease progression. Individuals who display a lower viral load during the first stage of the infection have higher chances to survive and control the infection [154] .
From the virus dynamics point of view, a successful vaccine is one that boosts the immune response, i.e., increases the parameter a', such that R 0 is reduced below the critical value of 1. Under such conditions, the infection will initially grow, but then viral load will rapidly decline and the infection will eventually be cleared from the organism. However, even if the vaccine induces an immune response that is not strong enough to reduce R 0 to a value below 1, it may still be beneficial. As described above, slowing down the initial exponential increase of the viral load has a negative effect on the viral set point. This type of imperfect vaccine is also called a therapeutic vaccine. In the case of HIV, it could help infected patients to control disease progression in two ways. First, it may reduce viral load during the chronic phase of the infection, and second it may reduce viremia below the threshold of inter-host virus transmission ( Figure 3 ) [137, 144, [155] [156] [157] .

Attenuation strategies and the evolution of virulence in RNA viruses

Individual viruses inside the quasispecies can display very different degrees of virulence. This trait is measured as the cell killing capacity, when considering cell culture experiments, and as the host morbidity or mortality, when focusing on whole-organism pathology [158, 159] . Replicative fitness is also highly variable among viruses of the same quasispecies, and in general, not directly correlated to virulence [158] . Understanding the evolution of these two viral traits is fundamental to understanding and controlling the spread of viral diseases and to the design of LAVs.

Conclusions

List of abbreviations used HIV: human immunodeficiency virus; HCV: hepatitis C virus; HVB: hepatitis B virus; RSV: respiratory syncytial virus; SIV: simian immunodeficiency virus; FMDV: foot-and-mouth disease virus; Ad5: adenovirus serotype 5 vector; LAV: live attenuated vaccine; IV: inactivated vaccine; NA: neuraminidase; HA: hemagglutinin; MHC: major histocompatibility complex; UPGMA: Unweighted Pair Group Method with Arithmetic mean; MRCA: most recent common ancestor; CTL: cytotoxic T lymphocyte; HLA: human leukocyte antigen; LD: linkage disequilibrium; PDNs: phylogenetic dependency networks; CBNs: conjunctive Bayesian networks; ODE: ordinary differential equations; HI: herd immunity.

Viral evolution

The evolutionary dynamics of viral infectious diseases can be analyzed at considerable detail today owing to advancements in high-throughput DNA sequencing technologies and statistical and computational modeling of these data [17] . Viral evolution occurs on different temporal and spatial scales and is shaped by different ecological processes within and between hosts. Integrated modeling efforts across these scales that make use of phylogenetic, population genetics, virus dynamics, and epidemiological methods are termed phylodynamics approaches [18] . Application of these techniques enabled the reconstruction of the molecular origin of the HIV pandemic [19, 20] and the explanation of influenza A epidemics by the interplay of natural selection and migration [21, 22] . Comparing viral DNA sequences is at the heart of the phylodynamics approach.

Vaccine development strategies

Recombinant and peptide-based vaccines. A great variety of vaccine strategies can be catalogued inside this category, most of them experimental. Recombinant vaccines are produced by the expression of a genetic construct that codifies viral peptides [47, 48] , subunits of the virus [49] , or whole viruses with genetic modifications including deletions of key proteins [50, 51] . Other strategies include live viral vectors that carry multiple copies of heterologous proteins of interest [52, 53] . Many of these strategies have failed, mainly because of the low immunogenic capacity of peptides or subunits, compared with the whole live or inactivated particle. The HBV vaccine is a yeast-derived recombinant vaccine. It contains the hepatitis B surface antigen which is one of the viral envelope proteins. HBV vaccine is the only recombinant vaccine currently approved and in use for humans [49] .

Phylogenetic methods in influenza vaccine design

Influenza infects large portions of its host population every season and immunized hosts are resistant to infection with the subtype they have been exposed to for several years. Therefore, selective pressure exists for the virus to diversify and to generate immunological escape variants. Indeed, the HA gene has been shown to be under strong selective pressure through immune surveillance [57] . Positive (diversifying) Darwinian selection acts at the antigen-determining sites of HA1, the most immunogenic part of HA. At these loci, significantly more non-synonymous than synonymous nucleotide substitutions are observed, and the rate of evolution is accelerated considerably as compared to other sites of the genome [58, 59] .
The evolutionary dynamics of influenza drive its immune escape and give rise to a new dominant strain every season. Therefore, vaccine design is not only supported by immunoinformatics methods for epitope prediction [63] [64] [65] , but also by statistical genetics and phylogenetic methods for analyzing genetic diversity and predicting evolutionary changes. To predict the evolution of the influenza HA gene, phylogenetic trees were constructed based on DNA sequences derived from viruses during the years 1983 through 1997 [22] . Eighteen codons were identified to be under positive selective pressure and the genetic diversity at these loci was significantly higher than at the other loci of the HA gene [59] . The rationale for predicting the next dominant virus is that extant strains with additional mutations at the 18 loci will be better adapted to evade the host immune response and thus have a selective advantage in the coming season. Phylogenetic analysis confirmed that the viral lineages with the greatest number of mutations in the positively selected codons were the ancestors of future H3 lineages in 9 out of 11 influenza seasons [22] . This approach to predicting the evolution of influenza relies on solving two classical evolutionary biology problems: the detection of genetic loci under selective pressure and the reconstruction of the evolutionary history of a set of individuals. Quantifying the relative contributions of selection versus random genetic drift is a longstanding task rooted in Kimura's theory of neutral evolution which predicts that most mutations are selectively neutral [66, 67] . Selection is typically identified by testing for departure from neutrality, although such deviations can also have different causes. The statistical tests are either based on the allelic distribution or on comparing variability in different classes of mutations [68] . Tajima's D is the prototype test of the first kind [69] . It detects differences in two distinct estimates of genetic diversity. The null distribution of the test statistic D is obtained from sampling genealogies according to the coalescent, a stochastic process describing the sampling variation [70, 71] . Similarly, the Ewens-Watterson test compares the observed to the expected homozygosity based on the Ewens sampling formula for the infinite-alleles model [72, 73] . In the second category of tests fall the McDonald-Kreitman test [74] and likelihood ratio tests based on the allelic distribution in nonsynonymous versus synonymous sites [58, 75] . Codon usage in influenza sequences has also been analyzed based on codon volatility, which measures the degree to which a random nucleotide mutation is expected to change the corresponding amino acid [76] .
Unlike distance-based methods, character-based methods follow character substitutions in the sequence explicitly. Maximum Parsimony is based on the minimum evolution principle and tries to find the tree that explains the data by the minimum number of mutations [83] . The method has been applied successfully to the analysis of influenza virus sequence data [84] . It is also computationally efficient, but lacks an explicit evolutionary model (other than minimum evolution), and it is not statistically consistent. On the other hand, probabilistic phylogenetic models have these desirable properties, but they come at a computational cost usually rendering exact maximum likelihood estimation impossible. However, approximate likelihood methods, such as DNAML [59] or QuartetPuzzling [85] , and methods making use of Bayesian inference, such as MrBayes [86] are applied in practice. For both distance-based methods and probabilistic phylogenetic methods, the choice of an appropriate model of nucleotide substitution is critical.
Immune responses to HIV infection vary depending on the genetic constellation of the human leukocyte antigen (HLA) locus. Antigen-specific T cell immunity is HLA-restricted and therefore mutations in HIV epitopes that allow escape from host immune responses are HLA allele-specific [119, 120] . Cytotoxic T lymphocyte (CTL) escape mutations have been shown to be stable under vertical transmission of the virus [121] . Thus, CTL escape presents an important driving force in shaping HIV evolutionary patterns. Differential HLArestricted viral evolution has been observed in several HIV-1 genes [122] . From a vaccine design point of view, it is pivotal to characterize CTL escape quantitatively in order to address limitations of immune stimulations and to minimize the risk of viral evolutionary escape. This task involves identification of HLA-specific escape mutations and prediction of mutational escape pathways.
Phylogenetic dependency networks (PDNs) are a class of probabilistic graphical models that account for these confounding factors [123] . PDNs explicitly represent selection pressure from multiple sources and model the dependency structure among escape mutations conditioned on a phylogenetic tree of the observed viruses and on the HLA types of their hosts. By analyzing 1000 individuals from a multicenter cohort, the statistical model identified a dense network of interactions between HLA alleles and HIV codons, as well as among HIV codons, reflecting the complexity but also the consistency of HIV adaptation to immune response [123] .
HIV mutational pathways have also been modeled in the context of evolutionary escape from the selective pressure of antiretroviral therapy. Several probabilistic models have been developed for describing the accumulation of amino acid changes in response to specific drugs or drug combinations, including Markov chains [124] , Bayesian networks [125] , mutagenetic trees [126] [127] [128] , and conjunctive Bayesian networks (CBNs) [129] . CBNs are a class of probabilistic graphical models that describe the order in which mutations occur. In this model, the partial order of mutations that maximizes the likelihood of the data can be learned efficiently from observed mutational patterns. CBNs allow for several escape paths with different probabilities and the partial order restricts the viral genotype space to the subset of those mutational patterns compatible with the partial order constraints (Figure 1 ). The rationale for this approach is that many combinations of mutations are never observed, for example, because they result in inviable viruses, or because they are too far away in sequence space from current strains to be reached on the relevant time scale. CBNs can explicitly represent the timeline of mutations occurring along escape pathways [130] and they have been extended to account for noisy observations [131] .
Computational models of viral escape dynamics have been applied successfully in the design of optimal combination therapies [132, 133] . Because the development of drug resistance is a major factor for treatment failure, not only the current resistance profile, but also the likelihood of evolving resistant viruses is a strong predictor of therapeutic outcome. The difficulty for the virus to escape from the applied selective drug pressure is known as the genetic barrier and it can be computed based on probabilistic models of accumulating mutations [127] . Retrospective analyses of large observational clinical databases have demonstrated that estimates of the genetic barrier based on viral progression models Figure 1 Conjunctive Bayesian networks describing HIV evolution under therapy with the two protease inhibitors ritonavir (A) and indinavir (B). The vertices of both graphs correspond to the same drug resistance-associated amino acid substitutions K20R, M36I, M46I, I54V, A71V, V82A, and I84V, in the HIV-1 protease, where K20R stands for a change from lysine (K) to arginine (R) at position 20, etc. Directed edges of the graphs denote partial order relations that constrain mutational pathways. An edge X Y indicates that mutation Y can only occur after mutation X has occurred. The H-CBN program from the CT-CBN software package [174] has been used to generate the models from 112 and 691 samples for ritonavir and indinavir, respectively. are independent predictors of treatment outcome. The genetic barrier improves therapy outcome predictions and the resulting models outperform standard-of-care expert rule-based treatment recommendations [134, 135] . Therefore, computational models of viral escape dynamics might also be useful for vaccine design. A successful HIV vaccine should not only minimize the distance to currently circulating strains, but also anticipate possible immune escape pathways of the virus. Although it is unlikely that the complete picture of escape pathways can be learned from data, improvements in terms of hindered and delayed escape might be possible, especially in the context of therapeutic vaccines where selective immune and drug pressure together may constrain virus evolution significantly and result in control of infection.
Besides genetic heterogeneity, the development of an efficient vaccine against HIV remains elusive because of the difficulties of inducing an efficient immune response [98, 136] . Individuals who control infection display a strong cellular response [137] [138] [139] . Many experimental vaccines have failed in directing the effector response to a more cellular profile [140] . Furthermore, HIV infection induces a low titer of neutralizing antibodies [141, 142] . Gp120 and gp41 are the HIV proteins exposed at the surface of the virion. These proteins are responsible for the attachment of HIV to the cell surface and the virus has developed several strategies to avoid recognition and blocking by antibodies. The region of the protein that interacts with the CD4 cellular receptor is a hypervariable loop. A great number of antibody-escape mutants are mapped to this region of the HIV genome. The loop however, is highly glycosylated and it is only exposed at the surface of the protein in the precise moment of the interaction with the cellular receptor [143] . These two combined features complicate the fitting of potentially neutralizing antibodies [142] .

Attenuation strategies and the evolution of virulence in RNA viruses

The experimental and theoretical results indicate that less virulent strains are more efficient in outcompeting the virulent ones in coinfected cells. Therefore, the fitness of variants of different virulence is density-dependent [168] . The cell competition model offers an explanation of several previous observations of suppression of high fitness mutants in dissimilar viral systems [169] [170] [171] [172] . This density-dependent selection due to varying efficiency of viral replication alone or in coinfection is reminiscent of the concept of a competition-colonization trade-off in ecology [173] . Here, virulent viral variants play the role of colonizers and viruses efficient within coinfected cells are competitors. The attenuating effect of competitor-colonizer competition appears even more pronounced if many viruses from a broad spectrum of virulence are considered.
44 section matches

Abstract

Twenty structures including 19 SARS-CoV-2 targets and 1 human target were built by homology modeling. Library of ZINC drug database, natural products, 78 anti-viral drugs were screened against these targets plus human ACE2. This study provides drug repositioning candidates and targets for further in vitro and in vivo studies of SARS-CoV-2. (Mengzhu Zheng), xingzhouli@aliyun.com (Xingzhou Li). † These authors made equal contributions to this work.
Abstract SARS-CoV-2 has caused tens of thousands of infections and more than one thousand deaths. There are currently no registered therapies for treating coronavirus infections. Because of time consuming process of new drug development, drug repositioning may be the only solution to the epidemic of sudden infectious diseases. We systematically analyzed all the proteins encoded by SARS-CoV-2 genes, compared them with proteins from other coronaviruses, predicted their structures, and built 19 structures that could be done by homology modeling. By performing target-based virtual ligand screening, a total of 21 targets (including two human targets) were screened against compound libraries including ZINC drug database and our own database of natural products. Structure and screening results of important targets such as 3-chymotrypsin-like protease (3CLpro), Spike, RNA-dependent RNA polymerase (RdRp), and papain like protease (PLpro) were discussed in detail. In addition, a database of 78 commonly used anti-viral drugs including those currently on the market and undergoing clinical trials for SARS-CoV-2 was constructed. Possible targets of these compounds and potential drugs acting on a certain target were predicted. This study will provide new lead compounds and targets for further in vitro and in vivo studies of SARS-CoV-2, new insights for those drugs currently ongoing clinical studies, and also possible new strategies for drug repositioning to treat SARS-CoV-2 infections.
Analysis of therapeutic targets for SARS-CoV-2 and discovery of potential drugs by computational methods, Acta Pharmaceutica Sinica B, https://doi.

. The Middle East Respiratory Syndrome Coronavirus

(MERS-CoV) broke out in the Arabian Peninsula in 2012 with a fatality rate of 35% 3, 4 . Both SARS-CoV and MERS-CoV are zoonotic viruses, and their hosts are bat/civet and dromedary, respectively 5, 6 . To date, no specific therapeutic drug or vaccine has been approved for the treatment of human coronavirus. Therefore, CoVs are considered to be a kind of viruses, of which the outbreak poses a huge threat to humans. Because Wuhan Viral Pneumonia cases were discovered at the end of 2019, the coronavirus was named as 2019 novel coronavirus or "2019-nCoV" by the World Health Organization (WHO) on January 12, 2020 7, 8 . Since 2019-nCoV is highly homologous with SARS-CoV, it is considered a close relative of SARS-CoV. The
International Virus Classification Commission (ICTV) classified 2019-nCoV as Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) on February 11, 2020. At the same time, WHO named the disease caused by 2019-nCoV as COVID-19. Common symptoms of a person infected with coronavirus include respiratory symptoms, fever, cough, shortness of breath, and dyspnea. In more severe cases, infection can cause pneumonia, severe acute respiratory syndrome, kidney failure, and even death. There is currently no specific medicine or treatment for diseases caused by SARS-CoV-2 9 .
CoVs are enveloped viruses with a positive RNA genome, belonging to the Coronaviridae family of the order Nidovirales, which are divided into four genera (α, β, γ, and δ). The SARS-CoV-2 belongs to the β genus.
In addition, viruses often bind to receptor proteins on the surface of cells in order to entering human cells, for example, the SARS virus binds to the angiotensin-converting enzyme 2 (ACE2) receptor [12] [13] [14] and the MERS binds to the DPP4 receptor 15, 16 . The therapies acting on the coronavirus itself include preventing the synthesis of viral RNA through acting on the genetic material of the virus, inhibiting virus replication through acting on critical enzymes of virus, and blocking the virus binding to human cell receptors or inhibiting the virus's self-assembly process through acting on some structural proteins.
For the development of medicines treating SARS-CoV-2, the fastest way is to find potential molecules from the marketed drugs. Once the efficacy is determined, it can be approved by the Green Channel or approved by the hospital ethics committee for rapid clinical treatment of patients. Herein, bioinformatics analysis on the proteins encoded by the novel coronavirus genes was systematically conducted, and the proteins of SARS-CoV-2 were compared with other coronaviruses, such as SARS-CoV and MERS-CoV. We conducted homology modeling to build all possible protein structures, including viral papain like protease (PLpro), main protease (3CLpro, also named 3-chymotrypsin-like protease), RNA-dependent RNA polymerase (RdRp), helicase, Spike, etc. Further, we used these proteins and human relative proteins [human ACE2 and type-II transmembrane serine protease (TMPRSS2) enzymes] as targets to screen ZINC U. S Food and Drug Administration (FDA)-approved drug database (ZINC drug database, ZDD), our own database of traditional Chinese medicine and natural products (including reported common anti-viral components from traditional Chinese medicine), and the database of commonly used anti-viral drugs (78 compounds) by virtual ligand screening method. This study predicts a variety of compounds that may inhibit novel coronaviruses and provides scientists with information on compounds that may be effective. Subsequent validation of anti-viral effects in vitro and in vivo will provide useful information for clinical treatment of novel coronavirus pneumonia.

Virtual ligand screening of potential drug targets of SARS-CoV-2

(1) some acting on enzymes or functional proteins that are critical to virus, preventing the virus RNA synthesis and replication; (2) some acting on structural proteins of virus, blocking virus from binding to human cell receptors, or inhibiting the virus's self-assembly process; (3) some producing virulence factor to restore host's innate immunity; (4) some acting on host's specific receptors or enzymes, preventing virus from entering into host's cells. The related target proteins include Nsp1, Nsp3 (Nsp3b, Nsp3c, PLpro, and Nsp3e), Based on homology models of the above 18 viral proteins (19 models) and 2 human targets, we resorted to structure-based virtual ligand screening method using ICM 3.7.3 modeling software (MolSoft LLC) to screen potential small-molecule compounds from a ZINC Drug Database (2924 compounds) and a small in-house database of traditional Chinese medicine and natural products (including reported common anti-viral components from traditional Chinese medicine) and derivatives (1066 compounds). Compounds with lower calculated binding energies (being expressed with scores and mfScores) are considered to have higher binding affinities with the target protein.

3C-like main protease (3CLpro)

It's worth mentioning, anti-asthmatic drug montelukast also showed low binding energy to 3CLpro. As shown in Fig. 5A , montelukast was well fitted into the active pocket of 3CLpro, in which lots of hydrophobic amino acids, just like Thr24, Leu27, His41, Phe140, Cys145, His163, Met165, Pro168 and His172 compose a relatively hydrophobic environment to contain the compound and stabilize its conformation. Hydrogen bonding was predicted between Asn142 and the carbonyl group of the compound (Fig. 5B ). DNA and RNA along the 5′-3′ direction in an NTP-dependent manner 37 . Importantly, it has been reported that the SARS-Nsp13 sequence is conserved and indispensable, and is a necessary component for the replication of coronavirus. Therefore, it has been identified as a target for anti-viral drug discovery, but there are few reports about Nsp13 inhibitors 38, 39 .
Besides the above targets, some non-structural proteins, including Nsp3b, Nsp3e, Nsp7_Nsp8 complex, Nsp9, Nsp10, Nsp14, Nsp15, and Nsp16, also play an important role in the virus RNA synthesis and replication, suggesting these proteins may be useful targets for the anti-viral drug discovery. The virtual screening results showed many anti-bacterial, anti-viral, or anti-inflammatory drugs from ZINC drug database and our in-house natural products/derivatives database displayed potential good affinity to these targets, and the detailed information of virtual screening results is shown in Supporting excel files (for ZDD screening results, file names as target.xlsx; for natural products screening results, file names as target_NP.xlsx).

Targets inhibiting virus structural proteins

Except for Spike protein, E protein (E-channel) possesses important biological functions for the structural integrity of coronavirus and host virulence. NRBD and CRBD of coronavirus N protein are needed for N proteins in host cells to bind with coronavirus RNA efficiently. Therefore, E protein or N protein (NRBD and CRBD domains) can be used as targets for the discovery of anti-viral drugs. Through virtual screening, many anti-bacterial, anti-viral, anti-tumor, anti-asthmatic, and anti-inflammatory drugs, etc. from ZINC database and our in-house natural products/derivatives database were found to display relatively good affinity to these targets. And the detailed results of virtual screening are given in Supplementary excel files.

Targets inhibiting virulence factor

There are three coronavirus virulence factors Nsp1, Nsp3c and ORF7a related to interfering host's innate immunity and assisting coronavirus immune escape. Nsp1 interacts with host 40S ribosomal subunit that induces specifically host mRNA degradation 42 and also inhibits type-I interferon production 43 . Nsp3c has ability to bind with host's ADP-ribose to help coronavirus resist host innate immunity 44 . Bone marrow matrix antigen 2 (BST-2) can inhibit the release of newly-assembled coronavirus from host cells. SARS-CoV ORF7a directly binds to BST-2 and inhibits its activity by blocking the glycosylation of BST-2 45 . These evidences suggest that Nsp1, Nsp3c and ORF7a may be potential targets for anti-viral drug discovery.

Virtual screening and target identification of common anti-viral drugs

The RdRp inhibitor GPC-N114 49 .The binding pocket of remdesivir was in the bottom of the RNA template channel, which position was for the acceptor template nucleotide ( Fig. 7A and B) . The compound was well fitted with the shape of the pocket, where it formed three hydrogen bonds with Asn497, Arg569 and Asp684.

Discussion

The triphosphate nucleotide product of remdesivir, remdesivir-TP, competes with RdRp for substrate ATP, so it can interfere with viral RNA synthesis. Our docking results show that remdesivir-TP binds to SARS-CoV-2 RdRp, with a score of -112.8, and the docking results are consistent with its original anti-viral mechanism, so we think remdesivir may be good in treating SARS-CoV-2 pneumonia. In addition, remdesivir also predicted to bind with the human TMPRSS2, a protein facilitating the virus infection, this is a new discovery and provides ideas for subsequent research.

Coronaviruses (CoVs) have caused a major outbreak of human fatal pneumonia since the beginning of the 21st century. Severe acute respiratory syndrome coronavirus (SARS-CoV) broke out and spread to five continents in 2003 with a lethal rate of 10% 1, 2

. The Middle East Respiratory Syndrome Coronavirus

Potential anti-coronavirus therapies can be divided into two categories depending on the target, one is acting on the human immune system or human cells, and the other is on coronavirus itself. In terms of the human immune system, the innate immune system response plays an important role in controlling the replication and infection of coronavirus, and interferon is expected to enhance the immune response 11 .
Blocking the signal pathways of human cells required for virus replication may show a certain anti-viral effect.
In the fight against coronavirus, scientists have come up with three strategies for developing new drugs 17 .
The first strategy is to test existing broad-spectrum anti-virals 18 . Interferons, ribavirin, and cyclophilin inhibitors used to treat coronavirus pneumonia fall into this category. The advantages of these therapies are that their metabolic characteristics, dosages used, potential efficacy and side effects are clear as they have been approved for treating viral infections. But the disadvantage is that these therapies are too "broad-spectrum" and cannot kill coronaviruses in a targeted manner, and their side effects should not be underestimated. The second strategy is to use existing molecular databases to screen for molecules that may have therapeutic effect on coronavirus 19, 20 . High-throughput screening makes this strategy possible, and new functions of many drug molecules can be found through this strategy, for example, the discovery of anti-HIV infection drug lopinavir/ritonavir. The third strategy is directly based on the genomic information and pathological characteristics of different coronaviruses to develop new targeted drugs from scratch.
Theoretically, the drugs found through these therapies would exhibit better anti-coronavirus effects, but the research procedure of new drug might cost several years, or even more than 10 years 11 .

Analysis, structure prediction and homology modeling of SARS-CoV-2 encoded proteins

We obtained the SARS-CoV-2 genome from Gene Bank. The genome sequence of Wuhan-Hu-1 was aligned with whole database using BLASTn to search for homology viral genomes. After phylogenetic analysis and sequence alignment of 23 coronaviruses from various species. We found three coronaviruses from bat (96%,