The Hidden War Inside Liver Cells: How Math Reveals HDV's Achilles' Heel

Roughly 12 million people worldwide carry a virus that most doctors rarely think about, yet it causes the most severe form of chronic viral hepatitis known to medicine. Hepatitis delta virus (HDV) can only infect people who already have hepatitis B — it is, in the most literal sense, a parasite of a parasite. But what HDV lacks in independence it more than makes up for in damage: co-infection is associated with a two- to six-fold higher risk of cirrhosis and liver cancer than hepatitis B alone.

For decades, the treatment options were almost nonexistent. Then came lonafarnib — a drug originally developed as a cancer therapy, repurposed because it blocks a molecular process HDV needs to assemble new viral particles. Lonafarnib doesn't touch the hepatitis B virus (HBV). That specificity turns out to be scientifically priceless. Because when you treat a co-infected patient with a drug that suppresses only one of two viruses living in the same liver cells, you get a rare natural experiment — and, if you're a mathematician, an extraordinary modelling opportunity.

That is exactly what Mhlanga, Shekhtman, Zakh, and colleagues exploited in this new study (Mhlanga et al., 2026). Using data from 15 co-infected patients enrolled in the LOWR HDV-1 clinical trial, they constructed and fitted a mathematical model of three interacting quantities — serum HDV RNA, serum HBV DNA, and hepatitis B surface antigen (HBsAg) — to extract biological constants that no clinical assay could measure directly. What they found forces a rethink of how these two viruses coexist, compete, and respond to treatment.

The Science

To understand the analysis, it helps to know what each of the three measured quantities represents. HDV RNA is the genetic material of hepatitis delta virus, circulating in blood — a direct readout of how much virus is being produced and shed. HBV DNA is the equivalent for hepatitis B. HBsAg — hepatitis B surface antigen — is a protein coat shed in vast excess by HBV-infected cells; it is so abundant in the blood of chronically infected patients that it has been used as a diagnostic marker for decades, and it is increasingly recognized as an important target in its own right because it can persist even when viral DNA is undetectable.

The 15 patients received one of three treatment regimens: lonafarnib (LNF) alone, LNF boosted with ritonavir (a pharmacological booster that raises LNF blood levels), or LNF combined with pegylated interferon-α (PEG-IFNα). Pegylated interferon-α — "pegylated" simply means chemically modified to last longer in the body — has immune-stimulating effects on top of its antiviral activity, which makes it a qualitatively different partner than ritonavir.

The mathematical model at the heart of the study is built around systems of ordinary differential equations — equations that describe how quantities change over time as a function of their current state. The framework tracks three linked processes: HDV-infected cells producing and clearing HDV RNA; HBV DNA dynamics that can be influenced by HDV activity; and HBsAg secreted by a pool of producing cells. Key parameters — including viral half-lives, treatment efficacy, and the strength of HDV's suppressive effect on HBV — were estimated by fitting the model to each patient's longitudinal data using nonlinear least-squares methods. The result is a set of biologically interpretable numbers with confidence intervals, not just curves that look right.

What They Found

The first result that jumps out is how swiftly lonafarnib acts on HDV, and how completely. After a short delay of zero to two days — probably reflecting the time needed for drug to accumulate in liver tissue — patients experienced a sharp first-phase decline in HDV RNA. Across all treatment arms, the model estimated that lonafarnib blocks 94% of HDV RNA production (95% confidence interval: 89%–97%). That is a strong initial effect, but it is not the whole story.

Lonafarnib Treatment Efficacy in Blocking HDV RNA Production

Label	Value
1st-Phase Efficacy (All Treatments)	94
Max Efficacy (2nd Phase)	98.9

Source: Mhlanga et al. (2026), LOWR HDV-1 study. 95% CI for first-phase: 89%–97%.

For patients on LNF monotherapy, the first-phase decline often flattened into what the authors call a "flat-partial-response" — a plateau where the virus stopped falling but didn't disappear. Many of these patients then experienced viral breakthrough (VB): HDV RNA stopped declining and began rising again, even while the patient was still on treatment. Viral breakthrough is a warning sign familiar from HIV and hepatitis C therapy — it usually signals that residual viral replication is sufficient to allow the emergence of resistance or immune escape.

Combination therapy told a different story. Patients receiving LNF with ritonavir or LNF with PEG-IFNα showed a biphasic HDV decline — that rapid first phase followed by a slower but continuing second phase, without viral breakthrough. The model explains the second phase as a time-dependent increase in treatment efficacy, deepening to a maximum of 98.9% inhibition. What drives that deepening isn't fully resolved, but in the PEG-IFNα arm it likely reflects the immune system being progressively activated to clear infected cells — a mechanism that operates on a different timescale than direct viral blockade.

HDV RNA Half-Life and HBV DNA Rebound

Label	Value
HDV RNA Half-Life (days)	1.26
Median HBV DNA Production Increase (fold)	4

HDV half-life 95% CI: 1.05–1.47 days. HBV rebound IQR: 1–28 fold. Source: Mhlanga et al. (2026).

The HDV RNA half-life — the time it takes for half of the circulating virus to clear, absent new production — was estimated at 1.26 days (95% CI: 1.05–1.47 days). That is fast. It means that if you could instantaneously stop all HDV production, serum levels would halve every 30 hours and drop by 99% in about nine days. The challenge, as the treatment data make clear, is that stopping production completely is hard.

Now for the twist. Lonafarnib does not target HBV. So in theory, HBV dynamics should be unaffected by LNF treatment — or at most passively reflect whatever happens when their shared host cells are perturbed. But that is not what happened. In every treatment arm except LNF+PEG-IFNα, at least one patient experienced a rise in HBV DNA while on therapy. The median increase in HBV DNA production rate, estimated by the model, was 4-fold (interquartile range: 1- to 28-fold).

Why would suppressing HDV cause HBV to rebound? The model's answer is elegant and, in retrospect, biologically plausible: HDV suppresses HBV replication when HDV levels are above a certain threshold, and when HDV falls below that threshold, the brake is released. In other words, HDV and HBV are not simply co-passengers in the same cell — they are engaged in an ongoing competition, and HDV has been winning. Remove HDV, and HBV surges to fill the space.

Treatment Arm Outcomes: HDV vs HBV Control

Label	Value
Biphasic HDV Decline	1
No Viral Breakthrough	1
No HBV Rebound	1
Rapid 1st-Phase Decline	3
HBsAg Stability	3

LNF = lonafarnib; RTV = ritonavir; PEG-IFNα = pegylated interferon-α. Source: Mhlanga et al. (2026). Scores: 1 = absent/unfavorable, 3 = present/favorable.

This is consistent with clinical observations that HDV co-infection tends to suppress HBV replication — patients co-infected with both viruses often have lower HBV DNA levels than those infected with HBV alone. The model now gives that observation a quantitative mechanism and a threshold structure: below a certain HDV level, HBV replication is no longer suppressed.

The one arm that escaped this pattern was LNF+PEG-IFNα. PEG-IFNα stimulates innate and adaptive immune responses — including interferon-stimulated genes that suppress HBV replication independently of HDV. The study suggests this immune activation may substitute for (or exceed) the suppression that HDV was providing, preventing the HBV rebound that occurred with LNF-only regimens.

HBsAg, the third quantity tracked, barely moved in any treatment arm. Surface antigen levels remained essentially flat throughout the study period. The model's explanation is structural: HBsAg is produced by a large, relatively stable pool of cells — including both actively infected hepatocytes and integrated HBV DNA in hepatocyte nuclei that continues secreting surface protein even without active viral replication. Because lonafarnib has no direct effect on HBsAg secretion and the cell pool turns over slowly, HBsAg is simply insensitive to short-term treatment changes. This finding has clinical implications: measuring HBsAg decline as an endpoint in lonafarnib trials may be uninformative unless trials are long enough to affect cell pool dynamics.

Why This Changes Things

The findings reframe HDV/HBV co-infection as a dynamic ecological relationship rather than two parallel infections (Mhlanga et al., 2026). The concept that HDV suppresses HBV is not new in the clinical literature, but having a mathematical model that quantifies the suppression threshold — and predicts the rebound — gives clinicians and drug developers a new lens through which to evaluate treatment designs.

The 4-fold median rebound in HBV DNA is not just academically interesting. Elevated HBV DNA is itself a risk factor for liver disease progression and hepatocellular carcinoma. A treatment that successfully suppresses HDV but simultaneously triggers a major HBV flare could, in theory, trade one danger for another. This may explain why clinical trials of HDV-specific therapies have sometimes produced unexpectedly mixed liver enzyme results — the liver is reacting to the HBV resurgence, not to the drug itself.

The finding that only LNF+PEG-IFNα prevented viral breakthrough in HDV and avoided the HBV rebound pattern has clear therapeutic implications. It suggests that the combination works not just additively but in a biologically complementary way: lonafarnib attacks HDV assembly while PEG-IFNα activates immune control over both viruses. This is a strong argument for combination therapy as the default design for future HDV trials, rather than using LNF as a monotherapy.

The estimated treatment efficacy of 94% for the first phase is encouraging. It places lonafarnib alongside other potent antivirals in terms of initial suppressive power. The issue is durability — and the model quantifies exactly where durability breaks down, which is the transition from first-phase to plateau in monotherapy patients. Knowing that the drug's efficacy can be deepened to 98.9% with the right combination partner is a roadmap for how to get from "good" to "potentially curative."

The HDV RNA half-life of 1.26 days is also a useful benchmark. It is shorter than that estimated for HCV in some studies, suggesting HDV particles are cleared from blood relatively quickly once production stops. This means treatment response, when it occurs, should be visible within days — and failure to see a rapid first-phase decline is itself diagnostic of poor drug penetration or very high viral production rates.

What's Next

The study has real limitations that the authors acknowledge. Fifteen patients is a small cohort. The confidence intervals on some individual-patient parameters are wide, reflecting the inherent challenge of fitting complex models to noisy clinical data with sparse time points. The model's representation of HBsAg dynamics is intentionally simplified — a constant producing-cell pool — and a more detailed model incorporating covalently closed circular DNA (cccDNA, the nuclear template that drives HBsAg production) would be needed to predict long-term surface antigen kinetics.

The mechanism behind the time-dependent increase in efficacy in the second phase also remains incompletely resolved. Is it pharmacokinetic — the drug accumulating further in liver tissue over weeks? Is it immune-mediated — infected cells being killed by an immune response that takes time to activate? Is it related to an intracellular accumulation of defective viral particles? The model identifies the phenomenon and quantifies its magnitude but cannot yet distinguish between these explanations from viral kinetics data alone.

Future work should test this model prospectively — using the estimated parameters to predict treatment response in new patients before the trial ends, as a true validation exercise. The model also opens the door to optimizing dosing schedules: if the threshold below which HDV releases its suppression of HBV can be estimated for individual patients, it might be possible to titrate lonafarnib dosing to keep HDV just above that threshold while an HBV-targeting drug is added to the regimen.

The broader ambition, implied but not fully stated in the paper, is a quantitative framework for combination cure strategies. HDV/HBV co-infection remains one of the most treatment-resistant forms of chronic viral hepatitis, and the path to functional cure — defined as sustained loss of HBsAg — will almost certainly require attacking both viruses simultaneously through complementary mechanisms. The model built by Mhlanga et al. is a step toward being able to simulate that process before running expensive and lengthy clinical trials.

For the 12 million people living with this dual infection, that ambition is not abstract. The mathematical curves in this paper represent trajectories of real liver cells, real viral particles, and real patients whose treatment choices will shape whether they develop cirrhosis in the next decade. Understanding the underlying dynamics — not just whether a drug works but how fast, how completely, and with what unintended consequences for the other virus — is exactly the kind of precision that distinguishes a treatment that buys time from one that achieves a cure.