The argument is not new in principle. More than two decades ago, researchers demonstrated that conventional endpoint assays provide cumulative measures of antiviral inhibition whose values depend on assay design and may bear limited correspondence to the instantaneous pharmacodynamic activity of a compound. What has changed is the quantitative rigor now brought to bear on the question. Recent mechanistic modeling of in vitro antiviral assays has shown that simply varying the sampling time can shift the estimated EC50 by more than fifteen-fold across the conditions examined. A compound that only modestly slows replication in the first cycle may look weak at 24 hours yet appear dramatically potent at 72 hours, because small per-cycle differences compound multiplicatively across successive rounds of infection.

Timing is only one axis of variability. The underlying kinetics of the virus-cell system—how fast cells are infected, how long the eclipse phase lasts, how rapidly progeny virions are produced—determine how strongly a given molecular inhibition is amplified at the population level. Modeling studies have shown that variation in infection rates alone can produce more than twenty-fold changes in apparent EC50, while variation in virus production rates can produce changes exceeding fifty-fold. When both parameters shift simultaneously, the difference in apparent potency can approach three orders of magnitude. Similar conclusions had been drawn earlier for HIV, where measured IC50 values were shown to depend not only on the intrinsic properties of the antiviral compound but also on the kinetics of the viral life cycle itself. The practical consequence is sobering: differences in EC50 between viral strains or variants do not necessarily indicate proportional differences in their molecular susceptibility to a drug.

The experimental context adds yet another layer. Multiplicity of infection, initial cell number, cell density, the timing of treatment relative to infection, the duration of drug exposure, and the number of replication cycles all influence the observed antiviral effect. Studies of influenza neuraminidase inhibitors, for example, showed that their activity decreases with increasing multiplicity of infection and depends strongly on when the drug is administered after infection. Cell type matters just as much, particularly for prodrugs that require intracellular activation. Comparisons of remdesivir and its parent nucleoside GS-441524 across Vero, Calu-3, and Caco-2 cells revealed substantial cell-type-dependent differences in antiviral activity, traced to differences in the intracellular formation of the active nucleoside triphosphate. An EC50 measured in one cell line therefore characterizes not only the drug and the virus but also the cellular system in which the drug is metabolized and the virus replicates.

Even the concentration-response relationship cannot be reduced to a single threshold. Two compounds with similar EC50 values may produce markedly different levels of inhibition at higher concentrations if their dose-response curves have different slopes. Work on antiretroviral drugs demonstrated that the slope of the dose-response curve differs systematically across HIV drug classes and strongly affects the inhibition achieved at clinically relevant concentrations. This observation motivated the concept of instantaneous inhibitory potential, a measure that incorporates both drug concentration and the slope of the dose-response curve. Reducing antiviral activity to a single concentration can thus obscure pharmacodynamic differences that matter enormously in the clinic.

Compounding these issues is the ambiguity of the term viral load itself. Infectious-virus titers measured by plaque assay, focus-forming assay, or the 50 percent tissue culture infectious dose method quantify replication-competent virus capable of initiating infection under specific assay conditions—not the total number of physical virions. Nucleic-acid-based assays such as RT-qPCR measure viral genome copies or transcripts regardless of whether those molecules reside within infectious particles. Extracellular RNA can originate from infectious virions, defective particles, or free material released during infection, while intracellular RNA reflects events inside infected cells, including incoming genomes, replication intermediates, or newly synthesized viral RNA. Cell-based measurements such as antigen staining, reporter viruses, and flow cytometry approximate the fraction of infected cells, whereas viability assays measure the remaining viable population without distinguishing infected from uninfected cells. These modalities are not interchangeable, and they can display different temporal behavior: in hospitalized influenza patients, infectious titers declined faster than viral RNA, and in SARS-CoV-2 infection, high RNA levels late in infection were a poor indicator of transmission potential.

To connect these disparate measurements to underlying biology, the field relies on mechanistic viral-dynamics models—typically systems of ordinary differential equations describing populations of uninfected cells, infected cells, and free virus. The classical target-cell-limited model, with parameters for infection rate, infected-cell loss, virus production, and clearance, remains the standard starting point. Its most influential application came from HIV research, where analysis of viral-load decline following antiretroviral treatment revealed the rapid turnover of plasma virus and enabled estimation of virion clearance rates, infected-cell lifespans, and viral generation time. Similar reasoning was later applied to hepatitis C, where models separated the direct inhibition of viral production by interferon from the slower loss of infected cells, and to hepatitis B following lamivudine administration. Extensions of the basic framework include eclipse phases representing cells infected but not yet producing virus, interferon-mediated antiviral responses, cytotoxic T cells, antibodies, and detailed intracellular models of genome replication, transcription, protein synthesis, and assembly.

Yet the power of these models is fundamentally constrained by what experiments can actually observe. Parameter identifiability—the question of whether model parameters can be uniquely recovered from data—imposes hard limits on mechanistic interpretation. Changes in infection rate, initial target-cell abundance, and virus production rate can compensate for one another and generate nearly indistinguishable viral-load trajectories. Adding biological complexity does not necessarily resolve this ambiguity; a more detailed model may fit the data better while producing less reliable estimates of individual processes. The most effective remedy is multimodal measurement: simultaneously quantifying infectious virus, total viral RNA, infected-cell fractions, and cell populations to constrain different aspects of the infection cycle. Landmark studies in SHIV infection demonstrated that joint fitting of such measurements allowed estimation of infected-cell half-life, infectious-virus half-life, burst sizes, and the ratio of infectious to total virions—quantities that could not be recovered from any single readout.

The observability problem intensifies as biological relevance increases. Animal models introduce systemic pharmacokinetics, tissue structure, and functional immunity, but many informative measurements require terminal sampling, reconstructing time courses from different animals rather than repeated observations of the same infection. Human infection offers the highest clinical relevance yet the least experimental control: the time of infection and effective inoculum are usually unknown, sampling often begins near or after the viral peak, and measurements are typically limited to nasal or nasopharyngeal swabs analyzed by RT-PCR. SARS-CoV-2 illustrated these constraints vividly. Viral RNA kinetics differ substantially between saliva and nasal swabs, with saliva loads frequently peaking earlier. A pooled analysis of over 2,000 upper-respiratory measurements from 605 individuals showed variation exceeding five orders of magnitude among individuals sampled at similar times after symptom onset. Moreover, viral RNA remained detectable for more than 19 days after symptom onset in one longitudinal study, while the median time to the first negative viral culture was approximately 11 days—persistent RNA detection does not necessarily indicate persistent production of replication-competent virus.

Despite these limitations, viral-dynamics modeling has delivered concrete value in antiviral pharmacology. Hepatitis C viral-kinetic models guided the development of interferon-based and direct-acting antiviral therapies, quantifying treatment effectiveness and predicting outcomes. A model integrating in vitro virology, pharmacokinetics, resistance, and early clinical viral kinetics for the protease inhibitor telaprevir prospectively predicted sustained virological response rates in later trials. For SARS-CoV-2, quantitative systems pharmacology models supported the selection of the nirmatrelvir/ritonavir 300/100 mg twice-daily regimen taken into pivotal trials, and population drug-disease models of oseltamivir informed dosing in immunocompromised adults, supporting early initiation and a longer 10-day course to reduce rebound risk. The review’s central message is ultimately one of disciplined interpretation: EC50 and EC90 remain indispensable for screening, but they must be read as properties of a drug-virus-cell-assay system rather than as context-independent constants. Antiviral effects are dynamic phenotypes, and understanding them requires the joint interpretation of measurements, models, and experimental design—an alignment that transforms a single number into a mechanistic statement about how a drug perturbs the viral life cycle.

Subject of Research: Mechanistic viral-dynamics modeling and the context-dependent interpretation of antiviral potency measurements

Article Title: Interpreting antiviral effects: from experimental measurements to mechanistic viral dynamics

Article References: Interpreting antiviral effects: from experimental measurements to mechanistic viral dynamics. (n.d.). https://doi.org/10.3389/fmolb.2026.1995001

Image Credits: AI Generated

DOI: 10.3389/fmolb.2026.1995001

Keywords: antiviral potency, EC50, viral dynamics, mechanistic modeling, parameter identifiability, SARS-CoV-2, influenza, HIV, hepatitis C, pharmacodynamics, in vitro models, viral load