CAR T-Cell and Endogenous T-Cell Interactions

Why Keep the Two T-Cell Populations Separate?

CAR T cells and endogenous tumor-reactive T cells can both contribute to tumor killing, but they do not recognize tumor in the same way. A CAR recognizes its target through the engineered receptor, whereas an endogenous T cell responds through its native antigen receptor and depends on antigen presentation. Tumor damage caused by CAR T cells may therefore influence endogenous activity indirectly: released tumor material can become available for presentation, and local inflammatory cues can support endogenous T-cell activation. These connections are not a reason to combine both populations into one generic effector compartment.

A useful QSP model preserves the two recognition routes and then represents the links between them. This makes it possible to ask whether a modeled CAR effect is direct killing, indirect recruitment or activation of endogenous cells, or both. The equations below are a deliberately compact representation of those mechanisms, not a claim that all tumors or CAR products behave identically.

A Compact Interaction Model

Assume that CAR T-cell abundance is specified over the period being modeled, or supplied by a separate CAR-cell model. This isolates the interactions of interest without re-deriving CAR activation or exhaustion. Let \(a_C\) and \(a_E\), each between zero and one, represent the fractions of tumor burden effectively recognizable by the CAR and endogenous populations. The saturable killing fluxes are

Here \(k_C\) and \(k_E\) are maximum killing rates per unit of the corresponding T-cell population, and \(H_C\) and \(H_E\) are half-saturation constants expressed in the same tumor-burden units as \(T\). These parameters need not be equal: the populations can differ in recognition, potency, and access to targets. If a population has no recognizable target (\(a_j=0\)), its corresponding killing flux is zero.

A minimal set of state equations is

The tumor equation includes logistic growth, with rate \(r\) and carrying capacity \(T_{\max}\), as well as both killing fluxes. The \(P\) equation is a lumped representation of tumor-derived antigen becoming available for presentation; \(\eta_C\) and \(\eta_E\) allow the two killing routes to contribute differently. It does not imply that CAR T cells themselves present antigen. The \(I\) equation represents a local inflammatory cue, with production coefficients \(\gamma_C\) and \(\gamma_E\) and clearance rate \(\delta_I\). The endogenous T-cell equation includes baseline input \(s_E\), loss at rate \(d_E\), and signal-dependent expansion or activation represented by \(r_E\). \(K_P\) and \(K_I\) are half-saturation constants for presentation and inflammatory signal.

The multiplicative signal terms express an assumption: in this illustrative model, endogenous expansion is strongest when both tumor-antigen presentation and local inflammatory support are present. They are phenomenological factors, not a detailed model of antigen-presenting cells, T-cell priming, trafficking, or cytokine networks. All rate constants and state units must be chosen consistently; for example, if time is measured in days, rates have units per day.

When Can Endogenous T Cells Expand?

The model gives a direct condition for the per-capita change in \(E\). First consider a period with \(s_E=0\). Dividing its equation by \(E>0\) gives

Thus, endogenous cells expand at that instant only if the signal-dependent expansion term exceeds their loss rate. CAR killing alone does not guarantee that condition: killing may produce little effective presentation, little local inflammatory signal, or both. Conversely, if \(s_E>0\), the total \(E\) can increase from input even when its per-capita net growth term is negative; the two contributions should not be confused.

A short derivation clarifies the link between killing and the two mediators. If \(K_C\) and \(K_E\) are approximately constant over a period, each mediator equation has the form \(dX/dt=Q-\delta X\), where \(Q\) is its production flux. At steady state, setting the derivative to zero gives \(X=Q/\delta\). Therefore,

These expressions are useful for examining the mechanism, but they are not a substitute for simulating transient dynamics when mediator clearance is slow or killing changes rapidly. They also show why antigen supply and inflammatory support should be represented separately: their source coefficients and clearance rates can differ.

Why More CAR Killing Does Not Necessarily Mean More Endogenous Activity

The coefficients linking tumor killing to presentation and inflammatory signal determine whether damage couples to endogenous-cell activity. They are not interchangeable. For example, an intervention or tumor context could permit substantial CAR-mediated killing while producing little of the modeled inflammatory cue. The following hypothetical comparison holds killing and antigen-presentation production fixed and changes only inflammatory-signal production.

Assumptions, Diagnostics, and Alternatives

The equations are most useful when their simplifying assumptions are visible. They assume that the chosen recognition fractions summarize target availability; each killing flux has a saturable response; antigen presentation and local signal can each be represented by one well-mixed state; and the signal-dependent term is an adequate summary of endogenous-cell activation over the modeled interval. The steady-state calculations further assume that killing fluxes are approximately constant long enough for mediator levels to approach steady state. When these conditions are not credible, simulate the mediator dynamics directly and consider richer states supported by data.

Useful checks include the following:

• Check state and flux units. Each killing flux must have units of tumor burden per time, and each source-minus-clearance equation must have units of its state per time.
• Check the no-recognition limits. Setting \(a_C=0\) must eliminate CAR-mediated killing, and setting \(a_E=0\) must eliminate endogenous-mediated killing.
• Check mediator behavior. With no production, positive \(P\) and \(I\) should decay under positive clearance rates. Under constant nonnegative production, the mediator steady states should be nonnegative.
• Check attribution against data. Tumor shrinkage alone may constrain total killing but may not identify how much came from each population or whether endogenous activity changed. Separate measurements or informative perturbations are needed to distinguish these mechanisms.

A common mistake is to add \(C\) and \(E\) into one effector pool and apply a single killing and activation rule. That may reproduce one tumor trajectory, but it removes the distinction between CAR recognition and endogenous antigen presentation and makes the indirect link from tumor damage to endogenous activity impossible to interpret. The correction is to retain separate states and killing fluxes, then specify which fluxes contribute to each mediator and which signals affect \(E\).

Compared with the compact ODE model here, a more detailed alternative could explicitly track antigen-presenting cells, peptide–MHC complexes, priming in lymphoid tissue, trafficking, or multiple endogenous T-cell states. Such detail is warranted when the data support it and the research question requires it; otherwise, extra states can be weakly identifiable. At the other extreme, a purely empirical model could let \(E\) change with time or tumor burden without an explicit antigen and signal pathway. That may be useful for prediction within a narrow setting, but it is less informative about the mechanism. No formulation removes the need to justify its recognition assumptions, parameter values, and observation model.

This tutorial treats \(I\) only as a local, lumped input to endogenous activity. It does not represent systemic inflammatory feedback, a detailed cytokine network, or a general account of clinical toxicity. Those mechanisms require additional states and evidence rather than being inferred from this local coupling term.