Strategies and efforts in circumventing the emergence of antiviral resistance against conventional antivirals

Abstract

Antiviral resistance stemming from rapid viral evolution and adaptation is a major challenge faced in treating viral infections. Here, we describe the mechanisms and factors underlying antiviral resistance and their implications to future drug development. Current improvements to conventional methods provide viable options to overcome antiviral resistance. Ongoing efforts in developing new antiviral strategies are also discussed. Examples from across virology are used to illustrate how virus evolution and antiviral therapy influence each other.

Introduction

The emergence of viral pathogens with pandemic potential has been a recurring, major concern for the global community^1,2. Through the relentless efforts of antiviral drug researchers and developers, numerous antivirals have been approved for use in treating infections caused by viruses such as the human immunodeficiency virus (HIV) and hepatitis C virus (HCV). However, despite these efforts, the increasing occurrence of antiviral resistance threatens to render existing antivirals ineffective^3,4. The recent COVID-19 pandemic has reiterated the importance of addressing this concern, with the development of effective antiviral therapeutics being outpaced by virus evolution⁴.

Conventional antivirals are known as direct acting antivirals (DAAs), which are designed to disrupt virus replication cycles by binding to specific viral proteins and inhibiting their activity, while minimising cytotoxic and off-target effects. DAAs target a variety of viral proteins essential for viral replication (Fig. 1 and Table 1). The first DAA discovered was idoxuridine, a nucleoside analogue used to treat herpes simplex keratitis⁵. Since then, antiviral development has come a long way. Many other drugs have been developed against viruses such as oseltamivir and zanamivir against neuraminidase (NA) of influenza virus⁶ or enfuvirtide against gp41 of HIV⁷.

Fig. 1: Types of antivirals and their mechanisms of action.

The various steps of the viral life cycle targeted by DAAs (purple) and HTAs (orange) are shown. A Viral binding inhibitors bind to viral glycoproteins, preventing them from binding host receptors, while host binding inhibitors act as host receptor antagonists, preventing their binding with viral glycoproteins required for viral entry. B Viral fusion inhibitors bind and disrupt the conformations of viral proteins responsible for mediating fusion between host and viral membranes. C Viral uncoating inhibitors bind to and interfere with viral proteins such as capsid or ion channels preventing capsid disassembly or endosomal acidification required for release of viral genome. D Viral nucleoside inhibitors bind to reverse transcriptases or polymerases and either cause chain termination or incorporation of a mutagen that drives a virus towards error catastrophe. E Host nucleotide depletion deprives viruses of purines and pyrimidines needed for genome replication. F Viral integrase strand transfer inhibitors bind to viral integrase active sites and displace viral DNA, preventing proviral incorporation into host genomes. G Host nuclear export disruption prevents the export of viral genomes and ribonucleoproteins out of the nucleus to their assembly site. H Host immune stimulants place host cells in an antiviral state by activating the expression of genes with antiviral activity against various steps of the viral life cycle. I Post-translational modifications are crucial for several viral proteins to function. HTAs that prevent such modifications could render these proteins non-functional, inhibiting the viral replication cycle at several different stages. J Viral assembly inhibitors interact with viral proteins, preventing their oligomerisation required for viral particle assembly. K An inhibitor that binds to and inhibits the function of a viral protein with multiple roles in the viral life cycle could simultaneously disrupt multiple stages at once. L Viral protease inhibitors bind to viral proteases and prevent the cleaving of viral polyproteins into functional proteins. M Viral egress inhibitors bind viral proteins required to separate the membranes of the host and new virions, thus preventing the budding of progeny viral particles from the host.

Table 1 Effects of various DAAs targeting different stages of the virus life cycle

A common problem with DAAs is their low genetic barrier to resistance. This genetic barrier refers to the mutational threshold a viral population needs to cross, beyond which clinically meaningful resistance develops⁸. The first orally administrated HCV protease inhibitors to receive United States Food and Drug Administration (U.S. FDA) approval were telaprevir and boceprevir^9,10. Due to their low potency, treatment had to be supplemented with pegylated interferon (IFN) and ribavirin. Both had issues such as poor safety profiles, which led to them being quickly replaced within 3 years of approval. Nevertheless, they paved the way for the development of antiviral therapeutics for HCV.

Aside from targeting viral proteins, antivirals can also be designed against host proteins, and are known as host-targeted antivirals (HTAs). HTAs target a variety of host proteins hijacked by the virus for replication (Fig. 1 and Table 2). One potential advantage of HTAs is that if they affect common pathways used by viruses of the same family, they have the potential to be broad spectrum. HTAs are theorised to have high genetic barriers to resistance because viral hosts are less prone to mutation. In some cases, becoming resistant to a HTA might require simultaneous mutations in several viral proteins, as was observed for the iminosugar UV-4B in dengue virus (DENV) and influenza viruses^11,12. Of the several HTAs that have been discovered in vitro, few have generated resistance mutations. Additionally, HTAs have been found to complement the action of DAAs and mutants raised against DAAs do not show resistance to HTAs^13,14.

Table 2 Effects of different HTAs on various host proteins hijacked by the virus for replication

In this review, we will discuss the development of antiviral therapeutics in light of virus adaptation and evolution. We will also review the current state of research in various alternatives to conventional antivirals.

Occurrence of resistance mutations to conventional antivirals

Treatment with an antiviral places a selection pressure on the viral population, causing a decrease in viral load. However, if complete viral suppression is not achieved, this creates a genetic bottleneck, from which drug-resistant variants are more likely to survive and replicate to replace susceptible viruses in the population^15,16.

Most viruses of concern are RNA viruses, that present several traits favourable to the establishment of resistance, including: poor replication fidelity, high replication rates¹⁷, genetic diversity¹⁸, genetic reassortment^19,20,21 and RNA recombination²². These factors drive high evolutionary rates and create ideal conditions for the emergence of resistant variants upon treatment with DAAs (Fig. 2A). Viruses are intracellular obligate parasites that hijack the host cell machinery to generate progeny. To disrupt viral life cycles, antivirals have to be present on the surface of or enter the cells themselves. Although DAAs are designed to target only viral proteins and not any of the host’s, early antivirals sometimes had the off-target effects of disrupting normal cellular processes whilst inhibiting viral life cycles. These undesirable side effects like cytotoxicity, restricted the dosage and frequency of treatment initially required to achieve viral suppression, and consequently provided an opportunity for the virus to develop resistance during suboptimal dosing.

Fig. 2: Characteristics of RNA viruses and emergence of antiviral resistance.

A Characteristics of RNA viruses that contribute to their high evolutionary rates and adaptability. B Comparison of direct and cross-resistance. Direct resistance is when a virus exposed to a single antiviral (Antiviral A) develops resistance to it. Cross-resistance is when a virus exposed to only one antiviral (Antiviral A) acquires mutations that confer resistance to it, as well as another antiviral (Antiviral B) to which the virus was not exposed. C Resistant mutants that confer a selective advantage over the wild type viruses in the presence of antivirals quickly become prevalent in the viral population. If these mutants are fitter than wild type even in the absence of antiviral, they are transmitted more often and becomes the dominant strain. Sustained transmission of these resistant viruses can render antivirals obsolete.

The first nucleoside reverse transcriptase inhibitor (NRTI) to be developed against HIV was zidovudine (AZT) which was clinically impactful in terms of disease progression and mortality, but suffered from a variety issues, most notably, cytotoxicity in the form of mitochondrial myopathy²³. AZT inhibited retroviral reverse transcriptase, but also mitochondrial DNA polymerase, causing a reduction of mitochondrial DNA over time with AZT therapy and clinically manifested as neuropathy, hepatic failure, and lactic acidosis in patients^24,25. Continuous monotherapy with AZT coupled with incomplete viral suppression led to the emergence of AZT resistance.

Following this, different classes of antivirals targeting the different stages of the retroviral life cycle were discovered. However, through a combination of viral, host and antiviral factors that limited the effectiveness of the therapy, complete viral suppression was not achieved. This led to the eventual emergence of resistance mutations against all classes of antivirals. In some instances, a single substitution alone was enough to confer a high level of resistance.

A single substitution of M184V in the reverse transcriptase of HIV-1 confers 300–600 fold reduction in potency in enzymatic and cell based assays to both emtricitabine (FTC) and lamivudine (3TC), NRTIs commonly used in anti-HIV therapy²⁶. In patients, resistance to the integrase inhibitor raltegravir can occur as three separate non-overlapping pathways with N155H, Q148R/H/K and Y143R/C substitutions in the catalytic core of HIV integrase, with each eventually leading to a high level of resistance^27,28. Lenacapavir, a HIV capsid inhibitor, had the potential to treat multidrug-resistant HIV that developed from other antiretroviral regimens, and was anticipated to have no pre-existing resistance mutations, given that it was the first of its novel class of inhibitors^29,30. However, several mutations discovered in vitro such as M66I, Q67H and N74D conferred over a 1000-fold reduction in susceptibility to lenacapavir²⁹. The N74D substitution which developed in as few as 3 weeks in patients demonstrated its low genetic barrier of resistance³⁰.

Direct resistance to a single antiviral drug is easily identified from screening patients with virological failure, which refers to the failure to keep viral load below a certain threshold³¹. Less commonly identified is the emergence of cross-resistance, which refers to a virus developing resistance to a drug it has never been exposed to, because of mutations that were selected for by a different drug (Fig. 2B). Ganciclovir-cidofovir cross-resistance during the treatment of cytomegalovirus (CMV) infection is one of the few well-described cases in the literature. Ganciclovir resistance mostly arises in the viral gene UL97, which codes for a viral protein kinase which converts ganciclovir into its active triphosphate form that stops the elongation of viral DNA by viral polymerase. However, during ganciclovir treatment, a rarer mutation, in the viral polymerase itself (UL54) confers ganciclovir–cidofovir cross-resistance³². Similarly in HIV, doravirine is a non-nucleoside reverse transcriptase inhibitor (NNRTI) that demonstrated efficacy against common NNRTI-associated mutations such as K103N, G190A and Y181C, at clinically relevant concentrations³³ and was evaluated for use as treatment for NNRTI-resistant HIV. However, mutations such as V106M or Y188L which confer resistance to doravirine can often be observed in patients that underwent first-line NNRTI regimens with drugs such as efavirenz or nevirapine³⁴, thus requiring genotypic antiviral resistance screening for patients with NNRTI resistance before administration of doravirine.

While any newly arising resistance mutation is a rare event, the speed at which it can accumulate and become fixed in a viral population varies greatly, and for some antiviral treatments can be a major concern. For instance, during the treatment of herpes simplex virus (HSV-1) with acyclovir and foscarnet, subpopulations which were resistant to both drugs became prevalent in the viral population in as little as 20 days^35,36. In HCV, a sustained virologic response refers to the inability to detect virus in blood for at least 24 weeks post-treatment, although an endpoint of 12 weeks post-treatment has become prevalent in most clinical studies³⁷. Failure to achieve this response is almost always accompanied by treatment-emergent drug resistance mutations, as demonstrated for chronic HCV patients undergoing grazoprevir therapy³⁸.

Once a resistance mutation to DAAs is established in an individual, there is risk that this mutation spreads across the population. Transmission of antiviral resistance mutations can occur both in the absence and presence of the antiviral treatment³⁹. Sustained transmission of such mutants, and genetic recombination with other virus strains can lead to fixing resistance mutations⁴⁰ rendering the antiviral treatment obsolete (Fig. 2C). This is the case for the influenza A virus (IAV) M2 ion proton channel inhibitors amantadine and rimantadine, to which most IAV has since become resistant worldwide. The S31N mutation in M2 of IAV confers high resistance at a low fitness cost, and became present in a majority of IAV cases, resulting in both drugs no longer being clinically used⁴¹.

Worse still, antiviral therapies may also have the unintended consequence of helping to drive the evolution of viruses to the benefit of the virus. Treatment of HCV resulted in R155K mutation in the NS3/4A protease which not only conferred resistance to multiple protease inhibitors, but also allowed the virus to escape from HLA-A*68-restricted CD8+ T cells⁴². Similarly, acyclovir treatment of herpes resulted in the emergence of non-cytopathic HSV-1 whose mutations also better evade the immune system⁴³.

Factors influencing the emergence of antiviral resistance

Antiviral-associated factors

There are several characteristics of an antiviral which contribute to how easily resistance develops, including genetic barriers, type of mutation, potency and duration of treatment, as well as pharmacokinetics/pharmacodynamics (Fig. 3). The genetic barrier to resistance of an antiviral refers to the number of mutations or ‘steps’ required to achieve resistance. Resistant variants requiring fewer mutations are more likely to occur. Additionally, the type of mutation also augments the genetic barrier, with transition mutations occurring more frequently than transversion mutations⁴⁴. Viral polymerases have a mutational bias towards transitions, since in the process of RNA elongation, it is easier to mis-incorporate a purine with another purine rather than a pyrimidine and vice versa. The effect of this mutational preference is that resistance mutations requiring transversions are less likely to occur than those requiring transitions⁴⁵.

Fig. 3: Virus, host, and antiviral-associated factors that contribute to antiviral resistance.

These different factors contribute to a suboptimal dose of antiviral being used during therapy, resulting in incomplete viral suppression which eventually leads to the emergence of antiviral resistance. A Virus-associated factors include reduced susceptibility to antivirals as a result of pre-treatment baseline resistance mutations, polymorphisms and variation in viral subtypes. B Host factors include genetic variation in hosts that affect their ability to respond to treatment or tolerate an antiviral. Various factors such as side effects, high pill burdens and frequent administration of antivirals contribute to poor adherence to antiviral regimens. C Antiviral factors include poor potency or the pharmacokinetic/pharmacodynamic profile of the antiviral that limits their effectiveness to suppress viruses. A low genetic barrier to resistance enables the viruses to gain resistance quickly.

At the same time, the potency of the antiviral also determines how much time the virus has to develop resistance. Highly potent antivirals sharply reduce the viral load, therefore usually only requiring a short exposure duration, minimising the probability for resistance mutations to develop. Low or intermediate potency drugs do not effectively decrease viral load and likely require a longer duration of treatment, allowing the virus more time to generate resistant mutants. Low potency together with a low genetic barrier to resistance provides the ideal conditions for the development of antiviral resistance. The relatively low potency and genetic barrier to resistance of 3TC is why up to 70% of patients with hepatitis B virus infections develop resistant variants after 5 years of treatment. Adefovir is also of relatively low potency but has a higher genetic barrier to resistance than 3TC and therefore only has 29% of patients with resistance after 5 years. Entecavir and tenofovir are considered high potency drugs with high genetic barriers, which is why only 1% of patients develop resistance after 5 years^46,47.

Highly potent antivirals might also develop resistance if they have poor pharmacokinetic and pharmacodynamic profiles. Many current antivirals such as ribavirin, adefovir, entecavir, acyclovir, AZT, 3TC and oseltamivir have hydroxyl groups in their structures which makes them polar and consequently, have poor permeability. On the other hand, drugs like chloroquine, lopinavir, arbidol, ritonavir, peramivir, baloxavir marboxil and letermovir have poor solubility due to the presence of hydrophobic groups in their structure⁴⁸. These limitations reduce the bioavailability of the antiviral at their target sites, such that they are below the concentration required for effective viral suppression.

Host-associated factors

The host receiving the antiviral also has an important role in the development of antiviral resistance. Even with the most potent of antivirals, adherence of the host during an antiviral treatment regimen often affects the outcome^49,50. For chronic infections such as HIV, a common problem is that the antiviral treatment has several issues such as high pill burdens, inconvenient drug administration, strict dieting, dose-limiting toxicity, and undesirable drug interactions, which result in suboptimal adherence (Fig. 3).

Interpatient variability in terms of genetics, particularly in immune-associated genes, also affects the effectiveness of treatment and the likelihood of antiviral resistance development. On human chromosome 19, single nucleotide polymorphisms (SNPs) in the IL28B gene which encodes for IFN-λ3, a type III IFN, were found to affect pegylated IFN/ribavirin treatment for HCV outcomes in patients coinfected with HCV/HIV. This information can be used to optimise treatment with IFN/ribavirin for those that respond well, while simultaneously minimising risk of exposure in patients that do not respond well⁵¹. Abacavir is a NRTI used to treat HIV, however, patients with HLA-B*5701 alleles tend to develop hypersensitivity from abacavir treatment⁵². Treating such patients with abacavir puts them at risk of poor adherence, and they should switch to an abacavir-free regimen to avoid development of abacavir resistance.

Environmental factors of the hosts also affect the adherence to an antiviral regimen. Surveillance studies on monkeypox virus-infected patient cohorts in Los Angeles reported up to 8.5% of patients studied were co-infected with HIV and exhibited low CD4 cell counts. The treatment of monkeypox virus infection typically involves oral tecovirimat, which requires high fat meals twice daily to ensure optimal absorption. Patient compliance to treatment regimen was affected by the poor’s accessibility to food, further highlighting the unlikelihood of meeting such dietary requirements in outbreak-prone regions, mostly in developing countries in Africa^53,54,55. Strict adherence to tecovirimat treatment may also be rendered unavailing or further exacerbated by pre-existing conditions, as was observed in a subset of the HIV-infected cohort that completion of a course of oral tecovirimat did not improve clinical outcome⁵⁵.In addition, recurrence of clinical symptoms is also possible and remains a concern even with good patient compliance, making extended therapy necessary^56,57,58.

Viral-associated factors

Baseline viral load and resistance

Patients with a higher baseline viral load, or copy number, are statistically shown to accumulate a higher absolute number of minority variants, which may confer antiviral resistance. In an Influenza Resistance Information Study, children aged 1–5 years old had higher viral loads and longer viral shedding compared to other groups of children aged 13 years and below in the study. These children also had the highest frequency of antiviral resistance mutations detected⁵⁹. A study to understand the factors contributing to failure to achieve SVR12 in HCV, found higher pretreatment viral loads increased the risk of treatment failure across different DAA regimens such as ledipasvir/sofosbuvir and sofosbuvir/velpatasvir⁶⁰.

In some viruses, certain genotypes or polymorphisms in the genomes confer a reduced susceptibility to antivirals, even if the host is treatment naive. These are referred to as baseline resistance associated mutations or polymorphisms (Fig. 3). This lowered susceptibility means that treatment doses that are effective against the usually susceptible virus are now sub-optimal, which can increase the likelihood of antiviral resistance emergence, since viral suppression is not achieved. Patients infected with subtype CRF02_AG/G HIV and a lack of known resistance-associated mutations, appeared to have lower baseline susceptibility to protease inhibitors such as lopinavir/ritonavir, compared to other HIV subtypes. Consequently, this increases the risk of resistance emerging should monotherapy be employed or a lack of adherence to antiviral regimen occur⁶¹.

Prior to treatment, HCV genomes containing resistance mutations are sometimes already present, although in relatively low abundance. These mutants could prevent complete viral suppression, allowing time for the virus to develop other mutations that further increase resistance⁶². Several baseline resistance-associated mutations and polymorphisms have been discovered in the different genotypes and subtypes of HCV which negatively affect treatment outcomes and contribute to the emergence of antiviral resistance and virological failure^63,64. The highly polymorphic residue 30 on NS5A of HCV has many substitutions across the different HCV genotypes which are associated with resistance to NS5A inhibitors. In HCV genotype 1a, glutamine is the prevalent residue, but substitutions with arginine or histidine residues occur naturally at a rate of 1%. In particular, Q30R substitutions confer a 1000-fold increase in resistance to daclatasvir⁶⁵.

The presence of pre-treatment resistance, as well as breakthrough mutations that emerge during treatment, resistance screening before and after (in the event of virological failure) is necessary in order to optimise treatment regimens⁶⁶.

Fitness costs

Aside from genetic barriers, viral fitness also determines whether the new genetic adaptations take hold in the population. While genetic barriers refer quantitatively to mutational steps required to achieve resistance, fitness refers qualitatively to mutants in terms of the selective advantage they bring to the virus under these growth conditions⁶⁷. Mutants with low genetic barriers are readily generated but are not always fit (Fig. 4A).

Fig. 4: Fitness costs and compensatory mutations.

A With multiple routes to develop resistance, both genetic barriers and fitness will determine which mutants will dominate. The higher the genetic barrier, the less likely resistance will emerge. Low genetic barrier and low fitness mutants (Mutant 1) are readily generated but selected against. Mutants with high fitness (Mutant 2), although unlikely to generate because of higher genetic barriers are likely to dominate the viral population if they emerge. Mutants with a very high genetic barrier (Mutant 3) are unlikely to appear because mutations incurring high fitness costs are required. B Resistance mutations confer a fitness advantage in the presence of an antiviral, but not in its absence. A double mutant with both resistance and compensatory mutations has better fitness than both wild type and resistant mutant alone in the presence of an antiviral, although not always in its absence. Compensatory mutations can occur stepwise with each compensatory mutant additively restoring the fitness lost by the resistance mutation. C Different mutations (Mutations A and B) work in synergy by conferring resistance to different antivirals (Antivirals 1 and 2). Synergistic mutations may also compensate for each other’s fitness cost (eg. low polymerase efficiency, low enzyme processivity etc.). D Compensatory mutations allow for the survival of resistance mutants which are otherwise non-viable by sufficiently restoring fitness (denoted by +) lost by the resistance mutation.

In general, once a virus is established in the human population, it reaches a local fitness maximum such that most mutations result in a fitness cost and very few would improve viral fitness any further. Thus, upon treatment with an antiviral, most resistant mutations will confer a fitness advantage in the presence of the drug; but will come with a trade-off fitness cost, in the original growth conditions without the drug. For example, the A335V mutation in non-structural protein (nsp)-13 helicase of coronaviruses confers partial resistance to remdesivir, at the cost of reduced helicase activity⁶⁸. These mutants would have a disadvantage compared to wild type in the absence of the antiviral⁶⁹, which explains the lack of transmission of these resistant viruses and why these mutants disappear upon cessation of treatment.

Under the selection pressure of an antiviral, if multiple routes to resistance are possible, the one with the least fitness costs, and thereby conferring the highest net fitness gain, will become dominant (Fig. 4A). As an example, both M184V and M184I substitutions in HIV reverse transcriptase confer resistance to both FTC and 3TC, with M184I usually preceding M184V. However, M184V has higher polymerase activity and will rise to become dominant over M184I over time^70,71.

In HSV infection, nucleoside analogues acyclovir and famciclovir are commonly used. These drugs target the viral polymerase, but require phosphorylation by the viral thymidine kinase (TK), which is encoded by the UL23 gene to become active. Mutations in UL23 can confer resistance to both drugs. This necessitates a switch to polymerase inhibitors that do not require phosphorylation by viral (TK), such as cidofovir and foscarnet. Mutations in the viral polymerase gene (UL30), however, confer resistance to all four drugs⁷². Despite this apparent advantage in broad resistance, most observed resistance mutations to acyclovir occur in UL23, suggesting these mutants bear lower fitness costs, rather than better fitness advantages. Indeed, HSV requires a functional DNA polymerase to replicate but can survive without TK under the right conditions, as evidenced by the existence of TK-negative and TK-low-producer mutant viruses⁷³. HSV also shows higher tolerance for mutations in UL23. Half of the mutants are insertions and deletions that render the entire enzyme non-functional. UL30, on the other hand, can only tolerate single nucleotide substitutions that occur in the domains of the polymerase responsible for proofreading and substrate binding, without rendering the enzyme non-functional⁷⁴. Nevertheless, the existence of HSV with mutations in both TK and DNA polymerase with similar replication fitness to wild type demonstrates the ability of viruses to alleviate fitness costs³⁶.

Compensatory mutations

Compensatory mutations are mutations that arise to alleviate or off-set the fitness costs incurred by mutations that confer antiviral resistance.

Compensation of the fitness cost of a primary mutation, by a subsequent secondary mutation, allows the double mutant to be fitter than both wild type and primary mutation viruses in the presence of the antiviral (Fig. 4B).

In influenza viruses, resistance to baloxavir, an inhibitor that targets the PA segment of the influenza replicase, is conferred by mutations I38L, I38T, or E199D for IAV and I38T for influenza B virus (IBV). These mutations result in lower polymerase activity and replication. Serial passage of these viruses revealed a D394N compensatory mutation for IAV and E329G for IBV that improved the replication of resistant viruses⁷⁵. If secondary mutations sufficiently compensate for fitness costs of the primary resistant mutation, the resulting double mutant may allow for the fixing and transmission of antiviral resistance in the viral population.

Resistance to the NA inhibitor oseltamivir is caused by the H274Y mutation in the NA gene, which comes at the cost of impeding proper folding of the enzyme, resulting in initial poor viral replication. However, secondary mutations restored viral fitness, enabling the H274Y mutant to become dominant and circulate⁷⁶. The L528M mutation in HBV not only restored replicative competence in drug resistant HBV, but further increased antiviral resistance⁷⁷. Compensatory mutations can sometimes work in synergy (Fig. 4C). E138K and K101E mutations in HIV RT which confer resistance to rilpivirine (RPV), have been demonstrated to compensate for reduction of replicative capacity and enzyme processivity (the ability of an enzyme to remain attached to the template when copying it) as a result of the M184I/V mutation, which confers FTC resistance. M184I in turn, compensates for the lower DNA polymerisation efficiency (the ability of the polymerase to generate products) caused by K101E^78,79.

Compensatory mutations appear after the antiviral resistance mutation, often in a stepwise manner on the way to their optimum (Fig. 4B). Mutations in the HBV virus reverse transcriptase M204V/I confer resistance to 3TC at the cost of lower in vitro replication efficiency. A L180M mutation emerges after M204V/I, partially restoring replication fitness back to around ~40% that of wild type, while also contributing to antiviral resistance. A third mutation V173L occurs following the first two, further increasing replication efficiency back to ~90% wild type levels⁸⁰.

Compensatory mutations augment the fitness of resistant viral mutants, allowing for pathways of escape which are normally selected against because of fitness costs. When designing broad spectrum antivirals, highly conserved regions of the viral genome are popular targets because they are present across several viral strains, genus and even families. These regions are conserved because they play essential roles in viral life cycles and any change to them comes at high fitness costs. One such example is the lysine residue in the F1 motif of polymerases in many single-stranded RNA viruses. This includes K229 in the PB1 subunit of influenza, K291 in nsp4 of chikungunya virus (CHIKV) and K159 in the RNA-dependent RNA polymerase (RdRp) of Coxsackie virus B3 (CVB3). Mutation of this in response to the antiviral favipiravir confers resistance with fitness costs of varying degrees (low in CHIKV, moderate in IAV and complete loss of viability in CVB3. As a result, respective compensatory mutations have emerged to offset these fitness costs (P653L mutation in the PA subunit of influenza polymerase, A239G in the RdRp of CVB3)^81,82,83 (Fig. 4D). Targeting of conserved sites in the viral genome places the virus in a difficult position of selecting between resistance to antiviral activity and impairment of function. However, the emergence of compensatory mutations in less conserved regions, offsets some of the fitness costs. This provides an escape for viruses in which they can keep their costly resistance mutations but with an overall smaller loss of fitness.

Understanding which mutations emerge over others, the genetic barriers to their development, the severity of their fitness costs and what compensatory mutations are possible could better inform on which compounds are worth pursuing. Along this thread, treatment with virus inhibitory peptide-based HIV entry inhibitors have been found to have a high genetic barrier to resistance and crossing that barrier required several mutations that came at high fitness costs in infectivity, fusogenicity, and replication, thereby supporting the development of this class of antivirals⁸⁴.

Existing strategies to overcome resistance to conventional antivirals

Combination therapeutics

Antivirals are often used in combination to prevent emergence of resistant variants of the virus, by increasing the genetic barrier to resistance or eliminating the infection before resistance can develop.

The long history of antiretroviral therapy provides many examples of this. As mentioned above, the first anti-HIV drug zidovudine was a NRTI and it was quickly followed by several other NRTIs: zalcitabine, didanosine, stavudine which all suffered from the same limitations because they shared a similar mechanism of action.

The subsequent discovery of two new classes of anti-HIV medications: protease inhibitors (PIs) and non-nucleoside reverse transcriptase inhibitors (NNRTIs), allowed for combination antiretroviral therapy (cART) with drugs of different classes, also known as highly active antiretroviral therapy (HAART). HAART significantly improved treatment outcomes in terms of morbidity and mortality. Other classes of HIV drugs were later discovered such as entry and integrase inhibitors, providing more options for HAART.

Combination therapy has also been applied to other viruses. Combination of daclatasvir with sofosbuvir and/or ribavirin have achieved high rates of sustained virologic response as the NS5A mutations do not affect sofosbuvir and ribavirin^85,86. A combination of tenofovir, sofosbuvir, daclatasvir, and ribavirin successfully treated HBV, HCV and hepatitis E virus⁸⁷.

Antiviral HTA and DAA combinations can display synergy particularly when they target different stages of the virus life cycle (Fig. 1). A combination of the mutagenic DAA molnupiravir (targeting viral replication), with TMPRSS2-inhibiting HTAs such as camostat, avoralstat, or nafamostat (targeting virus entry) synergistically suppressed SARS-CoV-2. Addition of brequinar, a HTA pyrimidine biosynthesis inhibitor, resulted in the three drugs further suppressing the virus⁸⁸. Combinations of a HTA inhibitor of the Raf/MEK/ERK cellular signalling pathway, ATR-002, used by many viruses during replication with molnupiravir, remdesivir or Paxlovid demonstrated synergy against SARS-CoV-2 in vitro⁸⁹. A combination of UV-4B and IFN against DENV were found to work in synergy. IFN stimulates the suppression of DENV in SK-N-MC (neuronal), HUH-7(liver) and HFF-1 (skin) cells, at therapeutically feasible concentrations; however, SK-N-MC cells are less responsive to IFN, possibly because they express fewer IFN receptors. On the other hand, UV-4B was cytotoxic in HUH-7 cells and cytostatic in HFF-1 cells at concentrations ≥100 µM. Combination of the two antivirals enabled complementary viral suppression across different tissue types at lower concentrations (Fig. 5A), which is particularly useful when treating pantropic viruses like DENV⁹⁰.

Fig. 5: Combination therapy, pharmaco-enhancements and lethal mutagenesis.

A The benefits of using antiviral combinations. Antiviral combinations can be used to exploit antagonistic resistance mutations which increase susceptibility to each other when they occur. Additionally, combinations lower doses of each constituent antiviral required for therapeutic effectiveness. Different antivirals (Antivirals A and B) also penetrate tissues to different degrees, which is especially important in pantropic viral infections. B Pharmaco-enhancements refer to modification of a drug to improve traits such as bioavailability and half-life. Boosting with a secondary antiviral (Antiviral B) slows the metabolism of the first antiviral (Antiviral A), prolonging exposure to the primary antiviral at the target site. C Mechanism of nucleotide analogue lethal mutagenesis. The analogue is incorporated into viral genomes during replication and increases the error rate of viral RNA-dependent RNA polymerase (RdRp), by allowing ambiguity in base pairing in subsequent rounds of replication, leading to an accumulation of erroneous mutations. Once the error rate exceeds the error threshold, fit viral genomes are no longer being replicated, resulting in viral extinction. D Resistance to mutagens can occur by several methods, such as mutations in the polymerase that prevent mutagen incorporation, mutations that increase fidelity to counter the increase in error rate, and the excision of mutagens by proofreaders which are present in some viruses. Mutagen efficacy is different for different viruses because of variation in mutational robustness which is the ability to express a phenotype despite deleterious mutations. Mutagens must be used at an optimal concentration, excessively high concentrations are genotoxic and toxic to the host, while sub-optimal concentrations may accelerate viral evolution.

Antiviral resistance mutations are sometimes antagonistic to each other, a feature which may be exploited in combination therapy. HCV A92K and P32del mutants demonstrated high resistance to ledipasvir, velpatasvir and elbasvir, but were susceptible to grazoprevir, ribavirin and sofosbuvir⁹¹. The K65R mutation in HIV is a common mutation that emerges during treatment with tenofovir, causing a clinically significant 2-fold reduction in susceptibility. However, when a K65R mutant gains an additional M184V substitution, which confers resistance to 3TC/FTC, it only had a 1.5-fold reduction in tenofovir susceptibility⁹². This further encourages the use of antiviral combinations, particularly with pairings that have resistance mutations that are antagonistic to one another (Fig. 5A). Drugs which were ineffective because of resistant mutations can be effective in combination⁹³. It should be noted that HTAs do not always prevent the emergence of resistance to DAAs when used in combination, as demonstrated by the emergence of oseltamivir-resistant influenza during treatment involving combinations of the HTA nitazoxanite with oseltamivir in vitro and in vivo¹³.

Pharmaco-enhancements

Some antivirals face the issue of limited bioavailability which refers to the extent of which the drug is in circulation inside the body and thereby accessible at the target site. This could be due to a number of reasons like poor solubility, quick metabolism or undesired drug interactions. One way to get around this problem is to modify the drug itself to have more favourable characteristics (Fig. 5B). The antiviral acyclovir is a viral DNA polymerase chain terminator and is the first line treatment for HSV encephalitis and has limited effectiveness against other DNA viruses like cytomegalovirus, Epstein Barr virus and varicella zoster virus. Acyclovir is limited by its poor bioavailability and short half-life, requiring frequent dosing to achieve therapeutic concentrations. The modification of acyclovir to make valacyclovir, a L-valyl ester from acyclovir improved bioavailability at least three times⁹⁴, allowing for a less frequent regimen which would improve adherence⁹⁵. Subsequent modifications to improve the half-life and spectrum of antiviral activity against other DNA viruses, resulted in the synthesis of ganciclovir, penciclovir and famciclovir^96,97,98,99.

If modification of a drug is not possible, or comes at the cost of reduced antiviral activity, another option is the addition of a pharmacokinetic enhancer. Pharmacokinetic enhancement, sometimes referred to as “boosting”, refers to the use of one drug to increase the effectiveness of another, the latter having antiviral activity (Fig. 5B). Early protease inhibitors against HIV such as saquinavir, indinavir, nelfinavir, and ritonavir were limited by their poor bioavailability, requirement of frequent dosing, high pill burdens and accompanying toxicity at therapeutic concentration. Ritonavir, although a protease inhibitor itself, was poorly tolerated in patients, but had the additional effect of being a potent inhibitor of cytochrome P450 CYP3A4 isoenzyme which is responsible for the metabolism of some other HIV protease inhibitors¹⁰⁰. This led to the development of ritonavir boosted protease inhibitors as a treatment, allowing for smaller doses (compared to monotherapeutic concentrations) of both ritonavir and the primary protease inhibitor administered, reducing toxicity and improving pill burden and adherence to the regimen¹⁰¹. Boosted protease inhibitors also achieved better viral suppression, reducing the chance of resistance emerging¹⁰². Which is why all protease inhibitors recommended for treating HIV are also recommended to be boosted with ritonavir, with lopinavir even being co-formulated with ritonavir as a pill (Meltrex)¹⁰⁰. In HCV ritonavir is part of the Viekira Pak regimen, a combination of ombitasvir a HCV NS5A inhibitor; paritaprevir, a HCV NS3/4A protease inhibitor; dasabuvir, a HCV non-nucleoside NS5B palm polymerase inhibitor and ritonavir co-packaged together. Ritonavir prolonging paritaprevir exposure in blood plasma¹⁰³.

Pharmacokinetic enhancement was also applied for COVID-19, such is the case of Paxlovid, which is the combination of the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) 3C-like protease inhibitor nirmatrelvir and ritonavir, which also reduced the metabolism of nirmatrelvir in the liver, allowing for prolonged nirmatrelvir antiviral effect¹⁰⁴ (Fig. 5B).

Another way of improving the pharmacokinetic profile of a drug is by changing the delivery system such as by delivery to the target site. Antivirals against respiratory infections such influenza, respiratory syncytial virus (RSV) and coronaviruses had better bioavailability at the target site, when delivered by inhalation through devices like nebulisers and aerosol inhalers, and also lower systematic exposure and toxicity¹⁰⁵. Also being explored are delivery methods like biocompatible targeted nanoparticles carrying drugs which have improved antiviral effect when compared to the same drug administered in vitro without nanoparticle delivery. Such delivery systems offer advantages of site specificity, controlled and sustained release, and compound stability. Additionally, hydrophobicity composition, size and surface to volume ratio of the particle can be modified to suit targeting needs, reducing off-target and undesirable side-effects, while improving bioavailability and lowering frequency of administration^106,107.

Lethal mutagenesis

Many rapidly replicating RNA viruses operate near an error threshold, which is a limit on the mutational rate that, if exceeded, results in the accumulation of deleterious mutations such that fit viral genomes cannot replicate in the viral population, therefore driving the virus to extinction¹⁰⁸. The strength of their high mutation rates in enabling rapid evolution and escape^109,110,111 could be exploited as a weakness using base analogues to cause lethal mutagenesis. When incorporated into viral RNA, antiviral nucleoside analogues, such as ribavirin and favipiravir, and more recently molnupiravir, allow ambiguous base pairing in subsequent rounds of replication, resulting in an accumulation of mutations in the viral population driving it towards non-viability (Fig. 5C).

Some viruses with high error rates have evolved a proof-reader which corrects replication errors. The RdRp of coronaviruses has the characteristically low fidelity of other RNA viruses, and despite their large genome size (~30 kb), replicates its genome with relatively high fidelity. This is due to the presence of nsp14, an enzyme that has the dual function of viral RNA cap methyltransferase and 3′ to 5′ exoribonuclease (ExoN) ability to excise exogenous nucleotides¹¹². This exonuclease activity, originally evolved to ensure sufficient fidelity for replication and mediate RNA recombination¹¹³, was even demonstrated to excise ribavirin from the viral genome (Fig. 5D), and is thought to explain the poor efficacy of ribavirin against coronaviruses in vivo¹¹⁴. Not all mutagens are excised. Favipiravir^115,116 and molnupiravir¹¹⁷ have been demonstrated to evade coronavirus ExoN proofreading function.

For viruses that do not have proofreading activity, mutagenic nucleotide analogues can be countered with mutations. A single G64S substitution in the polymerase of poliovirus conferred resistance to ribavirin by increasing the fidelity of the viral polymerase¹¹⁸. Although high fidelity polymerase poliovirus can replicate similarly to wild type viruses, the ability of the virus to adapt to new antiviral compounds or environmental changes is reduced. Prolonged ribavirin treatment of HCV in patients revealed several mutations in the NS5B polymerase that conferred resistance, possibly through improved replicative fidelity¹¹⁹. Another method of mutagen resistance is through specific discrimination against them by viral polymerases. FMDV polymerases containing the M296I mutation had a lowered capacity to incorporate ribavirin itself¹²⁰. In influenza, resistance to favipiravir is conferred by a K229R mutation in the PB1 segment of the polymerase. K229R does not increase fidelity, but reduces the incorporation of favipiravir (but not ribavirin) during RNA replication⁸¹ (Fig. 5D).

Despite these benefits, lethal mutagenesis has several limitations. Efficacy of the mutagen can depend on the mutational robustness of a virus which refers to the ability of the virus to maintain a phenotype while tolerating deleterious mutations¹²¹. For instance, CVB3 is less mutationally robust than poliovirus, resulting in the former being more susceptible to mutagens¹²². Treatment with mutagens requires high concentrations of drug, and intermediates of these drugs, may be utilised by host cellular processes, making them cytotoxic and or genotoxic to the host^123,124,125. On the flipside, sublethal doses of nucleoside analogues are thought to provide a narrow window for viral persistence and the eventual emergence of resistance^81,126. Worse still, these resistant viruses, coupled with their respective compensatory mutants, have the potential for transmission¹²⁷. Aside from resistance, if doses of the drug are not sufficient to clear the viral load, the drug could have the opposing action of accelerating viral evolution rather than viral extinction^14,128 (Fig. 5D).

Potential alternatives to conventional antivirals

Monoclonal antibodies

Monoclonal antibodies (mAbs) are mostly synthetic antibodies characterised by their monovalent affinity for a single epitope, rendering them an attractive tool as therapeutics¹²⁹. Since the advent of hybridoma technology, mAbs have hitherto been utilised as therapeutics for several diseases including cancer, neurodegenerative disorders and infectious diseases^{130,131,132,133,134}. Viral infections are no exception to this, with mAbs targeting viral glycoproteins (Fig. 6). Examples include nirsevimab, which targets respiratory syncytial virus (RSV) fusion protein and inmazeb (a mAb cocktail comprising atoltivimab, odesivimab and maftivimab), which targets the Zaire ebolavirus (EBOV) transmembrane glycoprotein. Both have already been approved by the U.S. FDA and are in active use¹³⁵.

Fig. 6: Potential alternatives to conventional antivirals.

Schematic depicting the general life cycle of a RNA or DNA virus, starting from viral attachment to release of new viruses (Steps 1 to 7). In RNA viruses, replication of their RNA genome occurs directly in the cytoplasm (Step 3a) whereas in DNA viruses, the DNA genome is first transported into the nucleus before being transcribed into mRNA (Step 3b). Figure shows the targets inhibited by each antiviral alternative, denoted by the red lines. A Monoclonal antibodies target the surface glycoproteins of viruses, preventing its binding to cell surface receptors and attachment of virus to the cell. B RNA interference is comprised mainly of siRNA and miRNA, each of which can be assembled with the RNA-induced silencing complex (RISC). siRNA associated with RISC induce gene silencing by targeting viral RNA from RNA virus genomes or the transcribed mRNA of DNA virus genomes. In contrast, miRNA associated with RISC repress the translation of RNA, thereby preventing the synthesis of any structural proteins required for the assembly of new viruses. C Defective viral genomes (DVGs) are typically present in RNA virus populations and bear mutations which can occur at any part of the genome. DVGs compete with wild type genomes for replication resources such as RNA-dependent RNA polymerase, or for newly synthesised structural proteins essential for virus assembly. D Locked-nucleic acids (LNAs) anti-sense oligonucleotides are comprised of nucleotide subunits modified by a methylene bridge within their ribose sugar LNAs can bind to specific, functional RNA motifs critical for viral replication.

Unfortunately, viruses with rapid evolution rates counter efforts in utilising mAbs as antivirals. This was evident during the recent COVID-19 pandemic, when casirivimab-imdevimab, previously effective against the SARS-CoV-2 Delta variant, had poor neutralising activity against the later Omicron and XBB variants^{136,137,138,139}. While casirivimab-imdevimab was initially effective against the Omicron variant during its inception, its efficacy waned against the later Omicron sub-variants, such as BA.2.75.2, BA.4.6 and BQ.1.1. This observation was consistent with other mAbs, such as sotrovimab and amubarvimab–romlusevimab¹⁴⁰. Similarly, It was reported that a substitution of glutamic acid on residue 484 of SARS-CoV-2 spike protein arose after treatment with bamlanivimab and casirivimab-imdevimab, reducing mAb efficacy^{141,142,143,144}. These prominent examples strongly suggest that the rise of immune escape mutants can be induced by mAbs.

How can we improve the efficacy of mAb therapy? Similar to DAAs, combinatory treatment with mAbs may present as an effective therapy option by widening the scope of epitopes that mAbs may target^145,146. mAb cocktails confer broad-spectrum antiviral effects either against closely related variants of the same virus or to viruses within the same family. For instance, a cocktail of two mAbs conferred protection against different variants of SARS-CoV-2¹⁴⁵, although further optimisation in vivo was suggested. In another example, two different EBOV-targeting mAbs derived from mice immunised with filovirus-based vaccines conferred protection to EBOV in mice and Sudan virus in guinea pigs. On the contrary, protection against EBOV in guinea pigs was not observed, suggesting that mAb cocktail efficacy may be dependent on the host species and adaptation of the virus¹⁴⁷.

In the past decade, advances in protein engineering allowed the use of bispecific monoclonal antibodies (BsAbs) in treating various diseases^148,149. BsAbs are specialised mAbs which target more than one epitope across different proteins, and thus might be a better alternative than conventional or combinatorial mAb therapy for viral infections^150,151,152. BsAbs may be designed to target two different viral structural proteins, as demonstrated by the work of Chang et al. and Moirangthem et al. in thrombocytopenia syndrome virus and IAV, respectively^153,154. In recent work on coronaviruses, BsAbs with other epitope-targeting configurations were employed as well. Lee et al. described the enhanced antiviral efficacy against Middle East respiratory syndrome coronavirus using a BsAb targeting two different epitopes on the spike protein surpassing that of parental mAb¹⁵⁵. In contrast, Li et al. reported a broad-spectrum antiviral approach against various beta-coronaviruses using BsAbs targeting beta-coronavirus spike protein and its respective receptor, angiotensin-converting enzyme 2 which is found in humans¹⁵⁶.

The abundance of epitopes that mAbs can target, coupled with the diversity of mAb-specificities which may be used in combination, positions mAbs as a strong class of antivirals. Despite the promising rudimentary studies, it should be noted that the generation of immune escape mutants may be closely associated with the use of mAb therapy. Resistance mutations have been reported to emerge in patients treated with mAb therapy¹⁵⁷. and continue to complicate its efficacy. It has been postulated that new variants emerged in patients treated with mAb therapy. Taking SARS-CoV-2 as an example, spike protein mutations at residues 455 and 456 reportedly arising from mAb-induced selective pressure were shown to have contributed to the total immune escape of the XBB.1.5.70 mutant from a panel of antibodies isolated from patients post-infection with the BA.4 or BA.5 strain¹⁵⁸. It remains crucial to be prudent in the use of mAb therapy to avoid generating novel immune escape mutants which may thwart the efficacies of existing antivirals.

RNA interference

RNA interference (RNAi) is a biological process present in mammalian and other invertebrate species, which induces gene silencing at the transcriptional or translational level¹⁵⁹. The two main arms of the RNAi process consist of small molecules known as small-interfering RNA (siRNA) and microRNA (miRNA). Their synthesis is mediated by the RNase III Dicer enzyme and the products associated with the Argonaute (AGO) family of proteins to initiate the assembly of the RNA-induced silencing complex (RISC) which gets involved in gene silencing^160,161 (Fig. 6). A third class of gene silencing RNA molecules known as piwi-interacting RNAs (piRNA) interact with the PIWI sub-group of the AGO family and are derived from Dicer-independent synthesis pathway^162,163. These different types of RNA molecules drive different pathways that achieve RNAi with slightly different mechanisms.

In the siRNA pathway, siRNAs target RNA molecules by binding to complementary bases. The RISC then cleaves the target RNA, causing degradation. In the context of viral infections, double-stranded RNA (dsRNA) molecules derived from the viral genome are recognised as pathogen-associated molecular patterns and processed by the Dicer enzyme into siRNA molecules which facilitate repression of viral replication¹⁶³. Upon successful loading into AGO, siRNA can bind to complementary virus-derived dsRNA strands which either exist as an inherent part of the viral genome or as intermediate by-products of viral replication. The binding mechanism driven by base-complementarity allows siRNA high target specificities to be designed as antivirals¹⁶⁴. This enables the development of siRNAs to target both DNA and RNA viruses via their transcribed mRNA and genomic RNA respectively. Theoretically, viruses may also be targeted at different steps of their life cycle, although more studies might be required to prove this speculation^165,166.

Several studies involving synthetic siRNA appear to point towards viral replication as the prime target for viral suppression. In herpes simplex virus type 1 (HSV-1), both Kalke et al. and Lasanen et al. repressed HSV-1 viral load and replication in vitro with a siRNA designed to target the HSV-1 UL29 gene, which encodes for a DNA-binding protein integral to the replication of HSV-1^167,168. The interference of HSV-1 replication was successful even in acyclovir-resistant strains. Similarly, Idris et al. reported significant antiviral activity, as well as prophylactic potential, when administering various siRNA targeting different SARS-Cov-2 helicase and RNA-dependent RNA polymerase, which are involved in viral replication¹⁶⁹. Likewise, Pei et al. have demonstrated the inhibitory effects of a recombinant siRNA targeting open-reading frame 7 (ORF7) of the varicella zoster virus (VZV) genome on virus replication¹⁷⁰. Interestingly, ORF7 has been suggested to be crucial for proper cytoplasmic envelopment in VZV¹⁷¹. This might open a window of opportunity to expand the repertoire of antiviral siRNAs, although more studies are necessary to establish whether it is feasible to target any other steps of the virus life cycle besides viral replication.

On the other hand, miRNAs are mostly host-derived, although they may also virus-derived. Most of the known virus-derived miRNAs are encoded by DNA viruses such as members of the Herpesviridae family, and to a lesser extent RNA viruses¹⁷². Unlike their RNA-cleaving counterparts, miRNAs repress RNAi by regulating the expression and translation of target genes. One of the most widely studied miRNAs is miR-155, which is expressed in response to infection by a diversity of viral pathogens including influenza, SARS-CoV-2 and enterovirus A71. Following its induction, miR-155 modulates both innate and adaptive immune responses^173,174.

Viruses have evolved to combat the antiviral effects of siRNA using viral suppressors of RNAi (VSR). VSRs were first described in plant viruses although in recent years, they have also been reported in several mammalian virus families^175,176,177. It is interesting to note that VSRs across genetically different virus families obstruct RNAi through the same mechanism, by competing with host Dicer enzyme for dsRNA as template, thereby preventing any downstream synthesis of antiviral siRNA¹⁷⁸. VSRs may also compete with existing siRNAs for cellular resources necessary for the assembly of RISC, as shown in the case of ZIKV where sub-genomic viral RNA can bind to the host proteins RHA and PACT and counteract RNAi¹⁷⁹. In influenza A virus, non-structural protein 1 (NS1) functions as a VSR and represses RNAi. Moreover, NS1 is highly mutable and this has been shown to give rise to phylogenetically different variants of IAV. It is unknown whether these different phenotypes might further diversify and complicate the current pool of RNAi suppression mechanisms^180,181.

In contrast, Lezcano et al. demonstrated that the 1A gene in drosophila C virus (DCV), which is typically involved in RNAi suppression, accumulated more polymorphisms in a RNAi-deficient infection model¹⁸². The team also reported increased viral fitness in DCV propagated in this model. Similarly, Samuel et al. reported increased virulence of Sindbis virus in Aedes aegypti bearing a defective Dicer-2 gene as compared to their wild type counterparts¹⁸³. Overall, these studies indicate how VSR evolution may occur constantly even in the absence of any selective pressure from the host RNAi response, against the backdrop of existing rapid viral evolution. The dynamics between viral evolution and host response remain complex and this warrants more studies on the modulatory potential of various RNAi pathways before exploring its potential for antiviral therapeutic development.

Defective viral genomes

Many RNA viruses are known to generate defective interfering particles (DIPs), which are typically characterised by their truncated genomes known as defective viral genomes (DVGs), due to their high mutation rates during error-prone viral replication¹⁸⁴. These truncated genomes may either work synergistically with wild type virus to facilitate the population’s adaptation, or antagonistically to hinder the life cycle of wild type virus when challenged with different environments^109,185,186. As the diversity of DVGs and mechanisms of generation have been reviewed in detail, this review will focus on the role of DIPs in the context of antiviral therapy¹⁸⁷.

Ever since their initial discovery in influenza viruses by Von Magnus, DIPs have been heavily described in experimental studies on influenza^188,189,190. DIPs have also been described in other major virus families as naturally-occurring particles during infection, such as in flaviviruses, paramyxoviruses and coronaviruses^191,192,193. Due to their smaller genomes, DIPs typically possess a competitive edge over wild type viruses within the same viral population during replication. Proposed mechanisms which confer fitness advantages to DIPs include competition with wild type virus for resources crucial for replication and assembly (Fig. 6), as well as induction of the host innate immune response^194,195.

DVGs with immunostimulatory potentials were previously described by Sun et al., who showed that DVGs generated during natural infection with RSV could inhibit viral replication in vivo¹⁹⁶. In contrast, the work conducted by Pelz et al. utilised cell lines lacking IFN activity against IAV replication. Instead, their results showed that DIPs which replicate faster exhibited stronger suppression of IAV replication in vitro, likely attributed to increased competition with wild type IAV due to its persistence¹⁹⁷.

In another recent study, authors demonstrated that HIV-derived therapeutic interfering particles (TIPs) engineered to replicate with a basic reproductive ratio (R₀) > 1, suppressed viral load in humanised mice 5 days post-challenge with HIV-1. An extended experiment involving infant rhesus macaque models challenged with chimeric simian/human immunodeficiency virus (SHIV) was also performed. Here, the authors reported that pre-treatment with a TIP adapted to Simian immunodeficiency virus reduced SHIV RNA and proviral DNA in lymphoid tissues, accompanied with enhanced survival rates. Reduction of viral loads in TIP-treated models was not attributed to immune responses due to the absence of elevated inflammatory response¹⁹⁸.

Currently, studies on the precise antiviral mechanism of DIPs are limited. Future work could aim to deepen the differential profiling of DVG sub-populations and characterise their phenotypes to elucidate the complex interactions between DVGs/DIPs, standard virus and host factors.

Targeting viral RNA secondary structures with locked nucleic acids

The genomes of RNA viruses are known to fold on themselves, forming what is known as secondary and tertiary RNA structures^199,200. These structures play important regulatory roles in the life cycle of many RNA viruses, for instance, the frame shifting element of coronaviruses that controls the essential programmed −1 ribosomal frameshift crucial for viral replication^201,202; the conserved cloverleaf-like RNA structure in enteroviruses that recruits proteins for genome replication²⁰³; some conserved RNA pseudoknot structures resistance to host Xrn1 enzyme degradation, responsible for the production of subgenomic flaviviral RNA that influences flaviviral pathogenicity^204,205,206; heterogeneous RNA conformations that control splicing and gene expression in HIV²⁰⁷ and influencing host immune suppression in influenza²⁰⁸. Conservation of these structures across viral genera and families^{203,205,206,209,210,211} and the important roles they play make these higher order RNA structures attractive targets when designing broad spectrum antivirals.

Locked nucleic acid (LNA) anti-sense oligonucleotides refer to a type of oligonucleotides that contain nucleotides modified by a methylene bridge within their ribose sugar that confers greater stability²¹². This stability enables LNAs to function as very potent compounds that bind with high affinity, favourable traits for therapeutic use²¹³. LNAs can be used to target functional RNA motifs critical for viral replication (Fig. 6), by binding to them and forming RNA duplexes that are degraded by RNase H, as demonstrated in Japanese encephalitis virus in vitro²¹⁴. LNA binding can also prevent the proper folding of RNA required for viral replication, as demonstrated in HCV where LNAs blocked conformational changes in the structure of NS5B CRE SL9266 pseudoknot, leading to impairment of viral translation²¹⁵.

In IAV, a RNA structure known as a packaging stem–loop 2 plays a role in packaging of virions in vitro and pathogenesis in vivo. Targeting this structure with a LNA ensured a 100% survival rate in mice challenged with a lethal inoculum of virus 2 weeks pre- and 3 days post-infection. These mice went on to develop an immune response robust enough to withstand 10-fold lethal inoculum re-challenge. Importantly, attempts at selection of resistance did not yield any mutants, indicating high genetic barrier to resistance, likely because the structure is critical for viability. This LNA sustained its potency even against IAV resistant to NA inhibitors, providing an option to treat IAV antiviral resistance. Applying the same approach to the conserved RNA structures of SARS-CoV-2 prophylactically, greatly reduced viral transmission in hamsters²¹⁰.

The binding of an LNA to the 5′ leader sequence of SARS-CoV-2, caused disruption of a stem-loop structure which was essential for replication in human cells in vitro and an in vivo mouse model. The LNA was effective against several SARS-CoV-2 variants of concern in vitro and when administered daily to mice intranasally²¹⁶. LNAs are flexible in the sense that they can be modified to target conserved RNA structures of other viruses which have similar essential roles in viral replication. Therefore, there is a need to comprehensively identify regions of viral RNA genome that are conserved and critical for replication beyond the proteins that they code for^217,218 and design LNAs that bind to and impair their function, as demonstrated for CHIKV where LNAs designed in against individual replicative elements in viral RNA were able to interfere with replication complex and prevent production of infectious progeny²¹⁹.

Current strategies to discover new antivirals

Deep mutational scanning

Deep mutational scanning (DMS) is a technique that entails the use of deep sequencing technology in conjunction with a mutant library of genes or genomes generated from reverse genetics to study the phenotypic effects of any mutation of interest. The output that can be derived from DMS has been used extensively in several applications across virology, including virus surveillance, virus evolution and prediction of immune escape mutants^{220,221,222,223}.

One of the major areas that is worth exploring with DMS is epistasis, which is characterised by the mutual influence of two or more mutations, which typically leads to an observable phenotypic change²²⁴. These fitness-associated interactions at the genetic level provide a wealth of information on the relationships between different viral genes or proteins and their impact on viral fitness, which may be exploited for the development of antiviral strategies²²⁵.

The effects of epistasis are commonly observed in highly mutable RNA viruses such as influenza viruses of the H3N2 subtype, which gain egg-adaptive mutations during the vaccine production process²²⁶. It was established that L194P, a leucine to proline substitution on residue 194 of hemagglutinin, exacerbates antigenicity and reduces vaccine efficacy. Interestingly, L194P bears opposing effects to several other amino acid substitutions such as G186V and this incompatibility yields detrimental effects to H3N2 influenza viability in vitro^227,228. Conversely, Lei et al. reported that double mutants bearing G186D and D190N substitutions were more prevalent in circulating H3N2 influenza strains and exhibited enhanced receptor-binding capabilities than their single mutant counterparts²²⁹. Such synergistic epistatic interactions have also been observed in NA substitutions, whereby circulating H1N1 strains bearing both the V241I and N369K mutations have increased prevalence and fitness²³⁰.

The consistent emergence of epistatic interactions are well-represented in SARS-CoV-2. One of the earliest mutations in the spike protein, D614G, was associated with increased infectivity and increased viral load. Virulence was further intensified with the emergence of the H655Y mutation in the gamma strain which also persisted in the later omicron strain. SARS-CoV-2 evolution remains an ongoing process, with constant emergence of new mutations further diversifying the combinatorial effects from epistatic interactions^231,232,233. Similarly, in HCV, epistatic interactions in the non-structural 3 (NS3) protein stabilised its conformation and promoted drug resistance even though most of the mutations alone were non-beneficial^234,235. These findings seem to suggest the favourable selection of viruses with strong epistatic interactions in nature.

The strong interplay between epistasis and virulence makes it an important factor when considering the design of antivirals. In recent years, studies utilised deep mutational scanning (DMS) to study resistance mutations in other pathogens such as bacteria and fungi in a bid to better understand their impact on antimicrobial development^236,237,238. Unsurprisingly, the use of DMS for such purposes has also gained traction in antiviral research in SARS-CoV-2 and IAV for example^233,239,240. Continued monitoring of resistance mutations is paramount for the exploitation of epistatic interactions in our favour.

Proteomics-based approaches

Protein-protein interactions (PPI) are ubiquitous and constitute an important part of cellular processes. In viral infections, virus proteins react with host proteins to carry out a variety of functions throughout the virus life cycle²⁴¹. These virus-host interactions form the basis for the design of HTAs, such as the HIV-1 antiviral maraviroc which prevents entry of HIV-1²⁴². Understanding these interactions can drive the discovery of new druggable targets.

Proteomics-based approaches such as mass spectrometry, coupled with advances in bioinformatics tools allowed us to identify targets within the host proteome^243,244. Kinases, which catalyse phosphorylation reactions across different biomolecules, is an example of a class of potentially druggable targets that has been studied extensively with proteomics-based approaches. These enzymes play a role in regulating viral replication^245,246. Haas et al. identified 13 kinases modulated by IAV from a pool of 332 IAV-host PPI and described two kinase inhibitors with pan-antiviral activity against both IAV and SARS-CoV-2²⁴⁷. Similarly, other studies utilising comparative proteomics to study phosphorylation profiles successfully identified kinase inhibitors with antiviral activity against enterovirus A71 and SARS-CoV-2 respectively^248,249,250.

The discovery of new PPI may also uncover the mechanisms of existing drugs and their potential to be repurposed as antivirals. This was evident during the COVID-19 pandemic, where Gordon et al. discovered 69 compounds that targeted a subset of the 332 high confidence PPI identified between SARS-CoV-2 and host proteins²⁵¹.

Utilising artificial intelligence

Artificial intelligence (AI) and its multitude of machine learning algorithms available, can be harnessed to reduce the costs of and accelerate the discovery of new antivirals²⁵². Machine learning approaches such as reinforcement learning utilise artificial neural networks trained with the appropriate data to generate novel compounds with the desired properties²⁵³. Deep reinforcement learning has been demonstrated to be able to design novel chemical entities against wild type and oseltamivir resistant influenza²⁵⁴ which showed antiviral activity in vitro and in vivo²⁵⁵.

AI can also assist in repurposing existing antivirals. Conventional drug-target prediction involves using available information in a “one-sided view”, but an approach called matrix completion with multi‐view side information (MVMC), utilises multiple available structural and chemical information about a given drug, as well as sequential or structural information of a target to complete an interaction matrix. MVMC demonstrated not only greater accuracy to previous methods of predictions, but also unveiled novel drug-target interactions for previously approved drugs²⁵⁶. Such approaches to drug repositioning during events like the COVID-19 pandemic, could assist clinicians in selecting prospective drug candidates to be further validated, streamlining the drug repurposing process in order to respond quickly to rapidly mutating viruses like SARS-CoV-2²⁵⁷. Versatility of such drug prediction tools allows application for many other human pathogenic viruses.

Conventionally, antiviral resistance is discovered through phenotypic and genotypic techniques, which are time consuming and costly. AI can be employed to study viral evolution²⁵⁸ and predict the emergence of antiviral resistance before they occur. The massive amount of sequencing performed on the genome of SARS-CoV-2 can be leveraged to study viral evolution by extensively mapping the effects of viral mutations on their fitness²⁵⁹. Mapping mutations in the viral genome regarding their benefits, fitness costs, and compensatory mutations, could reveal pathways of viral evolution and patterns of constraint which can guide antiviral drug design²⁶⁰. Although not as extensively done for SARS-CoV-2, mapping of fitness landscapes based on collective sequence data has been attempted (on the genome, but more often on specific genes) for many other pathogenic human viruses such as polio^261,262, HIV²⁶³, DENV²⁶⁴, influenza²⁶⁵ and HCV²⁶⁶.

Viruses are constantly evolving and as their genomes change, so do their fitness landscapes and even evolutionary constraints²⁶⁷. As such, there is a need to constantly build on and update previously generated maps, so that they provide updated information²⁶⁸. Recently, a new AI tool called EVEscape has been developed to study viral evolution, using a deep-learning approach and trained with available historical sequences and structural data, is shown to predict the potential of escape mutations to emerge²⁶⁹.

Discussion and concluding remarks

In this review, we outlined the current challenges that are imposed on conventional antivirals by virus evolution and the emergence of antiviral resistance. We also discussed the current alternatives which are being explored to DAAs. Amongst these alternative approaches, monoclonal antibodies and siRNAs are the only ones which have been implemented as therapeutics. Currently, there is a lack of information on proposed alternatives such as harnessing epistatic interactions, DIPs or LNAs. More studies are warranted to elucidate specific mechanisms and assess the safety of utilising these alternatives. Until then, we remain reliant mostly on existing DAAs as the main class of antivirals.

Despite the limitations of conventional antivirals and the challenges posed by viral evolution, new potential DAAs are continuously being discovered and developed. These drugs have greater potency, safety profile and specificity towards target tissues than their predecessors. Existing helicase-primase inhibitors like Pritelivir and Amenamevir are used to treat nucleoside analogue resistant HSV²⁷⁰, but lack sufficient penetration into neuronal tissue to affect viral latency²⁷¹. Newer helicase-primase inhibitors such as IM-250 have improved tissue exposure that allow for treatment of latent HSV and with low frequency of resistance²⁷². Nucleoside analogues and helicase inhibitors have been demonstrated to work synergistically against HSV²⁷³ providing more options for combination therapy. Nucleoside analogues against CMV often suffer from dose-limiting cytotoxicity. Filociclovir, a new nucleoside analogue has been demonstrated to not only have a better safety profile, but also be effective against some resistance mutants of previous nucleoside analogues²⁷⁴. A potential new combination treatment regimen against HIV, islatravir and lenacapavir with multiple mechanisms of action and target multiple stages of the life cycle, respectively, have high genetic barriers to resistance and demonstrated no antagonism nor cross resistance in vitro²⁷⁵.

We also discussed the potential of proteomics-based approaches and AI to study viral-host interactomes^{251,276,277,278}. Even though these approaches are new with rudimentary progress, they present opportunities to fill the current knowledge gaps in virus-host interactions for the discovery of new antivirals.