Latency, Reservoirs, and Viral Reactivation

Why Add a Latent Reservoir?

A within-host model can describe a rapid viral decline yet fail to explain why infection persists when measured virus is very low. One reason is that some infected cells may carry viral material without currently producing substantial virus. These cells form a latent reservoir: a population that can survive while productive infection is suppressed and can later contribute to renewed viral production through activation.

A reservoir is not the same thing as free virus below an assay’s detection limit. It is a distinct modeled population with its own entry, loss, and activation flows. The model therefore distinguishes persistence of latent infected cells from ongoing production by productively infected cells. This distinction is useful when interpreting suppression and rebound, but it also introduces parameters that may not be separately identifiable from plasma viral measurements alone.

Definition: A latent infected cell is an infected cell represented as not productively releasing virus under the model’s current state. A reservoir is the population of such cells. Activation is the transition of a latent cell into a productively infected state. A latent reservoir can persist even when free-virus concentration is low, but it contributes to productive infection only through modeled flows such as activation.

A Minimal Latent-Reservoir Model

Let \(T\) denote susceptible target cells, \(L\) latent infected cells, \(I\) productively infected cells, and \(V\) free virus. A basic model adds a latent pathway to a target-cell-limited infection model:

\begin{aligned} \frac{dT}{dt} &= s_T-d_TT-\beta TV,\\ \frac{dL}{dt} &= f\beta TV-(a+\delta_L)L,\\ \frac{dI}{dt} &= (1-f)\beta TV+aL-\delta_I I,\\ \frac{dV}{dt} &= pI-cV. \end{aligned}

Here, \(s_T\) is target-cell input, \(d_T\) is their per-capita loss rate, and \(\beta TV\) is the rate at which target cells become infected. The fraction \(f\) of new infections enters latency; the remaining fraction \(1-f\) enters the productive infected-cell compartment directly. Latent cells activate at per-capita rate \(a\) and are lost without activation at per-capita rate \(\delta_L\). Productively infected cells die at rate \(\delta_I\), produce virus at rate \(p\), and free virus is cleared at rate \(c\).

For example, if \(L\) is measured in cells and time in days, \(a\) and \(\delta_L\) have units day\(^{-1}\), and \(aL\) has units cells per day. The infection term’s units must also be cells per day; thus the units assigned to \(\beta\) depend on whether \(V\) is a total count or a concentration. All flows must have units matching the compartment whose amount is changing.

Model assumptions: The equations treat the populations as well-mixed and the listed rates as constant over the modeled interval. Infection is represented by mass action, the fraction \(f\) is constant, latent cells neither divide nor change activation rate with time, and each latent cell leaves the reservoir through activation or loss. These are simplifying assumptions, not universal biological properties. Treatment can be represented by changing relevant rates, such as reducing effective infection or viral production, but the assumed mechanism should be stated explicitly.

The latent-cell equation is a mass balance: new latent cells enter at rate \(f\beta TV\), while activation and non-activation loss remove cells at rates \(aL\) and \(\delta_LL\). Their sum is the total exit flow \((a+\delta_L)L\). This separation matters: increasing activation reduces the latent population but increases the immediate input to productive infection.

Reservoir Size, Activation, and Persistence

If latent-cell input is approximately constant at \(J_L\), the reservoir equation becomes \(dL/dt=J_L-(a+\delta_L)L\). At a steady state, input equals total exit, so the reservoir size and activation flow are:

L^*=\frac{J_L}{a+\delta_L}, \qquad J_{\mathrm{act}}^*=aL^* =\frac{aJ_L}{a+\delta_L}.

This result clarifies why reservoir size alone does not determine the rate at which latent cells reactivate. The activation flow depends on both \(L\) and \(a\). A large reservoir with a small activation rate can produce the same activation flow as a smaller reservoir with a larger activation rate. If new infection is completely blocked, then \(J_L=0\), and the reservoir declines rather than remaining at this steady state.

Worked Example: Latent Input and Steady Reservoir Size

Suppose a hypothetical model has 120 new infections per day, of which \(f=0.25\) enter latency. Let \(a=0.010\) day\(^{-1}\) and \(\delta_L=0.0020\) day\(^{-1}\). Assume these rates and the latent input remain constant long enough to use the steady-state calculation.

The latent input is \(J_L=f\beta TV=0.25(120)=30\) cells per day. The total per-cell exit rate is \(a+\delta_L=0.010+0.0020=0.0120\) day\(^{-1}\). Therefore,

L^*=\frac{30\ \text{cells/day}}{0.0120\ \text{day}^{-1}} =2500\ \text{cells}.

The activation flow is \(aL^*=0.010(2500)=25\) cells per day, and non-activation loss is \(\delta_LL^*=0.0020(2500)=5\) cells per day. The two exit flows sum to 30 cells per day, balancing the latent input. The remaining 90 of the 120 infections per day enter \(I\) directly under the model’s specified fraction \(1-f\).

When new latent-cell input stops, the model has a particularly useful solution. Write \(k_L=a+\delta_L\). For initial reservoir size \(L_0\), \(L(t)=L_0e^{-k_Lt}\). The reservoir half-life is the time for this quantity to fall to \(L_0/2\), namely \(\ln(2)/k_L\). The fraction of an individual latent cell’s eventual exit events that occur by activation is \(a/k_L\), provided \(a+\delta_L>0\) and the rates remain constant. The rest occur through non-activation loss.

Worked Example: Reservoir Decay Under Suppressed New Infection

Assume a hypothetical treatment fully blocks new latent-cell input at time zero. Let \(L_0=100{,}000\) cells, \(a=0.0020\) day\(^{-1}\), and \(\delta_L=0.00050\) day\(^{-1}\). These constant rates imply \(k_L=0.00250\) day\(^{-1}\).

The reservoir half-life is

t_{1/2}=\frac{\ln(2)}{0.00250\ \text{day}^{-1}} =277.3\ \text{days}.

After 30 days, the reservoir is \(L(30)=100{,}000e^{-0.00250(30)}=100{,}000e^{-0.075}=92{,}774\) cells, to the nearest cell. A substantial reservoir remains even though new infection has been stopped.

The cumulative number of activation events during those 30 days is the integral of \(aL(t)\):

\int_0^{30} aL(t)\,dt =\frac{aL_0}{k_L}\left(1-e^{-k_L(30)}\right) =\frac{0.0020(100{,}000)}{0.00250}(1-e^{-0.075}) \approx 5781\ \text{cells}.

This is an expected flow in the deterministic model, not a claim that a fractional or exactly specified number of individual cells can be observed. It describes how a slowly declining reservoir can continue supplying activated cells during suppression.

Activation as a Source of Productive Infection

Activation contributes \(aL\) cells per unit time to \(I\). If new direct infection is negligible and activation changes slowly relative to the turnover of \(I\), the productive-cell equation can be approximated as \(dI/dt\approx aL-\delta_I I\). Under a nearly constant \(L\), its quasi-steady value is \(I\approx aL/\delta_I\). If virus also changes rapidly toward its own quasi-steady state, \(dV/dt\approx 0\) gives \(V\approx pI/c\). These approximations require the source \(aL\) to vary slowly compared with the turnover rates of \(I\) and \(V\); they are not exact descriptions of every rebound trajectory.

Worked Example: From Activation Flow to Viral Production

Consider a hypothetical suppressed infection with \(L=50{,}000\) cells, \(a=0.0010\) day\(^{-1}\), and \(\delta_I=0.50\) day\(^{-1}\). Suppose activation is approximately constant over the time window of interest, direct infection is negligible, and productive cells and free virus have had time to approach their quasi-steady values.

The activation source is \(aL=0.0010(50{,}000)=50\) cells per day. Thus \(I\approx 50/0.50=100\) productive cells. Let effective viral production be \(p=100\) virions mL\(^{-1}\) cell\(^{-1}\) day\(^{-1}\), and clearance be \(c=10\) day\(^{-1}\). Then

V\approx\frac{pI}{c} =\frac{(100\ \text{virions mL}^{-1}\text{cell}^{-1}\text{day}^{-1})(100\ \text{cells})} {10\ \text{day}^{-1}} =1000\ \text{virions/mL}.

The calculation illustrates how activation can maintain a source of virus without new infection being the dominant source of latent cells. It does not establish that a measured viral concentration uniquely implies 50 activations per day: the result depends on the assumed reservoir size, activation rate, production rate, clearance rate, and quasi-steady approximation.

Stochastic Reactivation and Model Diagnostics

A deterministic differential equation describes average population flow. When a reservoir contains many cells, this can be a useful approximation. When it contains few cells, activation may occur as discrete, irregular events; a continuous flow such as \(aL\) can then conceal meaningful chance variation. For one latent cell with constant activation and non-activation loss rates, the probability of activation by time \(t\) is obtained by integrating its activation hazard over the probability that it has not yet exited:

P(\text{activation by }t) =\int_0^t a e^{-(a+\delta_L)u}\,du =\frac{a}{a+\delta_L}\left(1-e^{-(a+\delta_L)t}\right).

This expression assumes the two exit events are competing constant-rate processes. It is a probability for one cell; for many independent cells, the probability of at least one activation is calculated from the individual probabilities rather than by treating the expected number as a certainty.

Worked Example: Probability of Reactivation in a Small Reservoir

Suppose a hypothetical reservoir contains 1000 independent latent cells, each with \(a=0.0010\) day\(^{-1}\) and \(\delta_L=0.00020\) day\(^{-1}\). Calculate the probability that at least one cell activates within 30 days. Assume rates remain constant and each cell can activate at most once.

For one cell, \(a+\delta_L=0.00120\) day\(^{-1}\), so the activation probability is

q=\frac{0.0010}{0.00120} \left(1-e^{-0.00120(30)}\right) =\frac{5}{6}(1-e^{-0.036}) \approx 0.02947.

The probability that none of the 1000 cells activates is \((1-q)^{1000}\), by independence. Therefore, the probability of at least one activation is

1-(1-q)^{1000} =1-(1-0.02947)^{1000} \approx 1-1.0\times10^{-13}.

Under these assumptions, at least one activation within 30 days is overwhelmingly likely, even though the activation probability for any one cell is only about 2.95%. The expected number of activations is \(1000q\approx29.47\); this expectation and the probability of at least one event answer different questions.

Several diagnostics help determine whether the model is being used appropriately:

Check flow balance. Confirm that latent input is \(f\beta TV\), total exit is \((a+\delta_L)L\), and all terms in each equation have matching units.
Check assumptions against the setting. A fixed activation rate may be inadequate if activation changes with immune state, treatment, or time. Latent-cell proliferation or multiple reservoir types also require additional structure.
Check what the observations measure. Viral load measures free virus, not latent-cell count directly. A low or undetectable \(V\) does not by itself imply \(L=0\).
Check parameter trade-offs. If only viral measurements are available, a large \(L\) with small \(a\) may resemble a smaller \(L\) with larger \(a\), because both can generate similar \(aL\). Independent information on the reservoir or activation is needed to distinguish them.

The reservoir model adds a biologically interpretable source term to the simpler model without latency. Compared with an eclipse-phase model, which represents a delay between infection and productive output, a latent compartment represents a population that may remain nonproductive for a prolonged period and activate later. A single latent compartment is simpler to estimate and interpret, while multiple latent states or explicit stochastic models may better represent heterogeneous or very small reservoirs. Those additions need sufficient data to justify their extra parameters.

A common mistake is to conclude that a stable, low viral load proves the latent reservoir is large, or that a large reservoir must cause rapid rebound. Neither conclusion follows from reservoir size alone. In this model, the immediate activation source is \(aL\), while its downstream viral effect also depends on productive-cell loss, viral production, and viral clearance. Viral-load data alone can therefore leave reservoir size and activation rate confounded.

Key takeaway: Latency adds a reservoir with its own input, loss, and activation flows. Under constant rates, reservoir persistence depends on \(a+\delta_L\), while the immediate source of productive infection depends on \(aL\). Keep those quantities distinct when interpreting suppression or rebound.

Check Your Understanding

Work through each question, then compare with the solutions below.

A model has 80 new infections per day and \(f=0.10\). What is the latent input per day?
If \(a=0.0030\) day\(^{-1}\) and \(\delta_L=0.0010\) day\(^{-1}\), what is the reservoir half-life after new latent-cell input stops?
For \(L=20{,}000\) cells and \(a=0.00050\) day\(^{-1}\), calculate the activation flow.
Explain why an observed viral load alone may not identify both reservoir size and activation rate.

Solutions. (1) The latent input is \(f\) times the infection flow: \(0.10(80)=8\) cells per day. (2) The total reservoir exit rate is \(0.0030+0.0010=0.0040\) day\(^{-1}\), so \(t_{1/2}=\ln(2)/0.0040=173.3\) days. (3) The activation flow is \(aL=0.00050(20{,}000)=10\) cells per day. (4) The immediate input to productive infection depends on the product \(aL\). Different combinations of reservoir size and activation rate can give the same product, and viral load also depends on productive-cell turnover, virus production, and clearance.