Repeated Stimulation and CAR T-Cell Exhaustion

Repeated Encounters as a Dynamic Process

A CAR T cell can encounter antigen-positive tumor cells repeatedly. Each encounter may activate effector functions, while repeated or sustained stimulation can also contribute to a more persistent loss of functional capacity. The modeling challenge is to represent both the accumulated effect of stimulation and the recovery that can occur between encounters. A count of encounters alone does not capture their timing: the same number of contacts can produce different predicted states if one schedule allows more recovery.

The model below tracks an abstract exhaustion burden. It is a compact representation of state, not a claim that exhaustion is one directly measurable molecular quantity. The burden summarizes a persistent reduction in functional capacity attributable to repeated stimulation; it should be interpreted alongside measurements of killing, cytokine production, expansion, and cell-state markers. Acute activation during a contact is not identical to this persistent burden.

The encounter update and the recovery between encounters are represented separately. If \(E_n^-\) is the burden immediately before encounter \(n\), and \(E_n^+\) is the burden immediately after it, the update is

The increment is proportional to the remaining capacity \(1-E_n^-\), so the burden cannot exceed 1. Over a gap of length \(\Delta_n\), the exponential recovery assumption gives

For a constant interval \(\Delta\), define \(a=\exp(-k_{\mathrm{rec}}\Delta)\). Then \(0<a<1\), and the recurrence from one pre-encounter state to the next is

Assumptions and the Repeated-Encounter Steady State

This model is most useful as a transparent starting point under specific assumptions. Encounters are treated as equivalent: each has the same \(q\), and the interval is constant when a steady state is calculated. Recovery follows a single exponential rate, and no other process changes \(E\) between encounters. The functional relation \(C/C_0=1-E\) is a chosen linear approximation, not a universal biological law. In particular, this calculation does not imply that all CAR T cells exhaust at the same rate or that one phenotype marker uniquely measures \(E\).

At a repeated-schedule steady state, the burden immediately before an encounter is unchanged from one cycle to the next. Write this value as \(E_*^-\). Substituting the post-encounter update into the recovery equation gives

Rearranging yields the steady-state pre-encounter burden and then the post-encounter burden:

The denominator is positive because \(a(1-q)<1\). The equilibrium is stable: a difference \(d_n\) between two pre-encounter trajectories is multiplied each cycle by \(a(1-q)\), so \(d_{n+1}=a(1-q)d_n\). Since \(0\leq a(1-q)<1\), that difference shrinks over successive cycles.

The equations also make the timing effect explicit. For fixed \(q\), the steady pre-encounter burden as a function of recovery factor \(a\) has derivative

Thus, a larger \(a\)—less recovery between encounters—raises the steady burden. Because \(a=\exp(-k_{\mathrm{rec}}\Delta)\), increasing the interval \(\Delta\) lowers \(a\) and lowers the predicted steady burden, all else equal. This is a model-based comparison, not a conclusion that longer intervals are always clinically preferable: tumor growth, cell trafficking, and other biological processes are outside this calculation.

Finite Schedules and Unequal Gaps

A steady state is not necessary to reason about a finite sequence of contacts. For any gap \(\Delta_n\), use \(a_n=\exp(-k_{\mathrm{rec}}\Delta_n)\). Starting from \(E_n^-\), apply the encounter update and then multiply by \(a_n\) to obtain the next pre-encounter burden. This stepwise calculation preserves the actual schedule rather than replacing it with an average interval.

When intervals vary, two schedules with the same number of encounters need not have the same burden at every time point. Closely spaced contacts can produce a higher peak burden than contacts separated by longer gaps. The sequence of interval-specific recovery factors matters, especially when the question concerns function immediately before or after a particular encounter. Replacing variable gaps by one average interval is an approximation and should be checked against the actual timing pattern.

Diagnostics, Alternatives, and Limits

A useful model should be confronted with observations that can challenge its assumptions. In a controlled repeated-stimulation experiment, one could measure function after a standardized challenge at multiple times, alongside viable CAR T-cell counts and selected cell-state markers. If functional capacity recovers at a different rate after stimulation is withdrawn than the model predicts, a single exponential recovery term may be inadequate. If repeated equal contacts produce progressively different increments, a constant \(q\) is also suspect. Marker changes can support a biological interpretation, but a marker alone does not establish the functional state represented by \(E\).

• Check timing: record the encounter or stimulation schedule and the times at which function is measured. Sparse measurements may miss transient declines or recovery.
• Check the state mapping: compare predicted \(1-E\) with functional readouts. A curved or threshold-like relationship may fit better than the assumed linear mapping.
• Check the encounter assumption: differences in antigen exposure or signaling strength can make contacts non-equivalent. In that case, a common \(q\) may conceal important variation.
• Check what is being measured: cell loss, altered cell composition, and reversible functional suppression can all lower a bulk assay, but they need not be the same mechanism as the modeled exhaustion burden.

The most important practical limitation is parameter identifiability. With only one endpoint after repeated stimulation, a large burden per encounter and slow recovery can both explain reduced function. Time-course data across several schedules are more informative because the interval dependence constrains recovery separately from the encounter effect. Even then, this simple model is not a substitute for a richer state-transition model if distinct progenitor, effector, or memory populations are central to the question.

A continuous-input model is one alternative when stimulation is better represented as persistent exposure rather than discrete contacts. For example, a rate equation can balance stimulation-driven burden formation against recovery. That approach is useful for sustained exposure, but it can obscure the consequences of individual encounter timing. Conversely, the discrete recurrence used here is convenient for pulse-like contacts but compresses the biology into one state and two parameters. The choice should follow the temporal structure of the data and the intended prediction.

A common mistake is to conclude that two schedules with the same number of contacts must have the same effect, or to use \(q\) as if it were a fixed amount added at every event. In this model the increment is \(q(1-E^-)\), so it depends on the burden-free fraction remaining; recovery also depends on the gap. The correct comparison applies the update and recovery equations in schedule order, and states clearly that the result is conditional on the model assumptions.