Biphasic Viral Decline and Long-Lived Infected Cells

Why Viral Decline Can Have Two Phases

After an effective antiviral intervention, viral load may decline rapidly at first and then more slowly. One mechanistic explanation is that the intervention greatly reduces the generation of newly infected cells, while already infected cells continue to produce virus as they are lost. If the productively infected population contains groups with different lifespans, the short-lived group can dominate the early decline and the longer-lived group the later decline.

This is a mechanistic interpretation, not a conclusion that follows from curve shape alone. A biphasic curve is compatible with multiple processes; here the model isolates how different infected-cell survival rates and production rates can produce two phases. The earlier material on early viral decline slopes distinguishes observed slopes from the biological quantities they may reflect.

A Two-Population Model

Let \(I_f(t)\) and \(I_s(t)\) denote short-lived and long-lived productively infected cells per millilitre at time \(t\), respectively. Let \(V(t)\) be free-virus concentration in copies per millilitre. The loss rates of the two infected-cell populations are \(\delta_f\) and \(\delta_s\), in inverse days, with \(\delta_f>\delta_s>0\). Each cell population produces virus at constant rates \(p_f\) and \(p_s\), in copies per cell per day. Free virus is cleared at rate \(c\), in inverse days.

The interpretation requires specific assumptions. Over the interval being modelled, new infection is negligible, for example because treatment has strongly suppressed infection of additional target cells. The infected-cell loss and production rates and the free-virus clearance rate are constant. The two infected-cell groups are treated as distinct, well-mixed populations, and there are no additional changing sources or sinks of virus in this model. These assumptions are simplifications, not universal properties of treated infections.

Solving the first two equations gives \(I_f(t)=I_f(0)e^{-\delta_f t}\) and \(I_s(t)=I_s(0)e^{-\delta_s t}\). Thus, the production from each group declines at that group’s cell-loss rate, provided its per-cell production remains constant. The virus concentration responds to both changing production and viral clearance.

From Infected-Cell Survival to Viral Load

The exact solution for free virus, when \(c\ne\delta_f\) and \(c\ne\delta_s\), is

This expression shows that viral clearance can contribute a transient with rate \(c\). If clearance is much faster than the infected-cell changes, that transient typically fades quickly. After it has faded, the viral concentration is often well approximated by the sum of two exponentials:

This approximation requires \(c\) to be substantially larger than both cell-loss rates and the observation interval to extend beyond the initial viral-clearance transient. The amplitudes \(A_f\) and \(A_s\), in copies per millilitre, reflect both the initial number of cells and their per-cell production. They are not simply cell counts.

Worked Examples

Why the Apparent Slope Changes

The total log-slope is the negative weighted average of the two cell-loss rates, with weights equal to the components’ current contributions to viral load. Early in the decline, the fast component may dominate, so the total slope is close to \(-\delta_f\). As that component disappears more quickly, the slow component can dominate and the slope approaches \(-\delta_s\). At the exact crossover, the components have equal load, so the total log-slope is \(-(\delta_f+\delta_s)/2\).

The crossover time exists at a positive value when the initial fast amplitude exceeds the slow amplitude and \(\delta_f>\delta_s\). In general it is given by

A small slow amplitude can postpone the crossover substantially or place it beyond the observation period. Conversely, a sizeable slow amplitude can make the slower component apparent relatively early. The word “phase” describes the dominant contribution over an interval; it does not mean the system switches abruptly from one cell population to another.

Diagnostics, Alternatives, and Limitations

A two-component model is most credible when the data cover enough time to show a reproducible change in slope, measurements are sufficiently frequent and precise to distinguish that change from noise, and the assumed treatment interval is consistent with strongly suppressed new infection. On a log-scale plot, a single exponential appears approximately straight; systematic curvature toward a shallower late slope is consistent with a second, slower component. It is not proof of that mechanism.

Compare the principal modelling choices:

A common failure mode is to infer the lifespan of a particular cell population directly from one observed slope. The approximation links component slopes to cell-loss rates only if the assumptions hold and that component dominates the measured load. The observed total slope may be between rates, and a rapid viral-clearance transient can further affect early observations.

Another limitation is identifiability. In the approximate model, viral-load data primarily inform the amplitudes \(A_f\) and \(A_s\), not \(I_f(0)\) and \(p_f\) separately, or \(I_s(0)\) and \(p_s\) separately. A larger cell population with lower production per cell can yield the same amplitude as a smaller population with higher production. Additional measurements or justified constraints are needed to separate those quantities.

The two-population model also assumes constant production and cell-loss rates. If those rates change, or if the infection is not sufficiently suppressed to make new infection negligible, a two-exponential fit may summarize the curve without accurately describing its mechanism. The relevant diagnostic is not just whether a fitted curve looks smooth, but whether its assumptions and predicted component behaviour are credible for the interval studied.