Cytotoxic T-Cell Recognition and Tumor-Cell Killing

From Contact to Tumor-Cell Death

A cytotoxic T cell can only kill a tumor cell it encounters, but an encounter is not itself a death event. The T cell must recognize the presented antigen, and recognition must lead to lethal execution rather than disengagement or an ineffective contact. A useful QSP model therefore separates three processes: encounter frequency, recognition probability, and the probability of killing after recognition.

This separation matters when interpreting model parameters or comparing interventions. A higher tumor-cell death rate could reflect more frequent contacts, a greater fraction of contacts recognized, or a greater chance of lethal execution once recognition occurs. Tumor-reactive T-cell states, antigen presentation, and T-cell function provide biological context for these terms, but a measured death rate alone does not identify which process changed.

The encounter coefficient summarizes how efficiently effectors and targets come into contact in the modeled compartment. It is not a probability: \(aE\) is a rate. In contrast, \(q\) and \(r\) are dimensionless probabilities between zero and one. In this formulation, \(q\) represents antigen-specific recognition during an encounter, while \(r=P(\text{lethal execution}\mid\text{recognition})\). Thus their product \(qr\) is the probability that an encounter produces a lethal event, under the model's assumptions.

The basic formulation treats \(E\) as the available effector concentration and assumes target cells are sufficiently similar to share the same encounter and conditional probabilities. It also assumes that contact events can be represented by a mass-action rate, and that the parameters are constant over the time interval being considered. These are modeling assumptions, not biological laws. If effectors become unavailable during prolonged contacts, or if target cells differ substantially in antigen presentation, the simple formulation may not represent the process adequately.

Deriving an Effective Death Hazard

Suppose that, for one tumor cell, potential encounters occur as a Poisson process with constant rate \(\lambda_{\mathrm{enc}}=aE\). Each encounter independently produces a lethal event with probability \(qr\). The probability of no lethal event in a short interval \(\Delta t\) is approximately \(1-\lambda_{\mathrm{enc}}qr\Delta t\), ignoring terms that become negligible as \(\Delta t\) approaches zero. It follows that the instantaneous hazard of lethal events is \(\lambda_{\mathrm{enc}}qr\).

For a homogeneous population of \(X\) viable tumor cells, the corresponding CTL-mediated loss flux is \(h_{\mathrm{CTL}}X\), in cells/mL/day. If tumor-cell growth and other death processes are also represented by rates \(g\) and \(d_0\), both in day\(^{-1}\), a simple net balance is:

This balance assigns CTL-mediated loss to a separate term; it does not imply that CTLs are the only influence on tumor burden. It also treats \(E\), \(q\), \(r\), and \(a\) as fixed over the interval. If those quantities vary, then the hazard varies over time. For a time-varying hazard and no cell division or other death, the surviving fraction over an interval from 0 to \(t\) is \(\exp\left(-\int_0^t h_{\mathrm{CTL}}(u)\,du\right)\). A constant hazard gives the familiar surviving fraction \(\exp(-h_{\mathrm{CTL}}t)\).

Representing Recognition and Execution as Separate Steps

The single probability \(r\) can be kept as a conditional probability, or it can be linked to a simple model of the recognized contact. For example, suppose that after recognition a tumor-cell–T-cell conjugate can either proceed to lethal execution at rate \(k_{\mathrm{kill}}\), or dissociate without killing at rate \(k_{\mathrm{off}}\). Assume these are constant competing exponential rates, with no other possible outcome during the contact.

The probability of lethal execution before dissociation is then the probability that the execution event occurs first. The two event times are independent exponential waiting times, so the probability that execution occurs first is the execution rate divided by the sum of the two rates:

Both \(k_{\mathrm{kill}}\) and \(k_{\mathrm{off}}\) have units day\(^{-1}\), so \(r\) is dimensionless. Increasing \(k_{\mathrm{kill}}\) raises the probability of lethal execution; increasing \(k_{\mathrm{off}}\) lowers it. This expression concerns the outcome of a recognized contact. It does not increase the number of encounters, and it does not by itself describe how often recognition occurs.

This competing-outcomes representation assumes that a contact has just become recognized, the two specified outcomes are the relevant alternatives, and the rates remain constant during that contact. If contacts have multiple stages or additional outcomes, the conditional probability must reflect those stages. A model with an explicit conjugate state can also represent a delay between recognition and tumor-cell death. The compact hazard \(aEqr\) is most useful when contact dynamics can be summarized by an effective probability of lethal outcome per encounter and do not substantially restrict subsequent encounters.

What a Change in Killing Rate Can—and Cannot—Tell You

The product form makes the model's central interpretive limitation clear: a death hazard combines multiple processes. A larger \(h_{\mathrm{CTL}}\) does not, by itself, show whether encounters became more frequent, recognition became more likely, or recognized contacts became more lethal. Independent information about effectors, antigen recognition, or contact outcomes is needed to assign a change to one mechanism.

For contrast, consider two hypothetical interventions with the same encounter rate of \(0.0500\) day\(^{-1}\). In condition A, \(q=0.40\) and \(r=0.50\), giving \(h_{\mathrm{CTL}}=0.0500(0.40)(0.50)=0.0100\) day\(^{-1}\). In condition B, \(q=0.80\) and \(r=0.50\), giving \(0.0200\) day\(^{-1}\): the increase comes from recognition probability. If instead \(q=0.40\) and \(r=0.90\), the hazard is \(0.0180\) day\(^{-1}\): the increase comes from conditional execution. These calculations isolate mechanisms for illustration; in real systems, more than one component may change at once.

Assumptions, Diagnostics, and Alternative Formulations

Before interpreting \(aEqr\) as a mechanism, check that the assumptions match the data and question. The basic model assumes that encounter frequency is proportional to effector concentration, each modeled encounter has the same recognition and execution probabilities, and lethal encounters can be represented as independent events. It also assumes that the relevant antigen is presented and that the modeled target population is adequately represented by a shared set of parameters. The model does not account for changes in effector function or target properties unless those changes are explicitly allowed to alter its parameters.

Useful diagnostics ask whether the model's separate pieces are compatible with observations. Compare its predicted encounter frequency with contact or conjugate measurements, its recognition probability with antigen-specific recognition measurements, and its lethal outcome probability with target-cell death following recognized contacts. A predicted death rate that is too high despite plausible encounter and recognition measurements may point to an overestimated execution probability, an unmodeled delay, or an incorrect contact approximation. Conversely, matching tumor-cell loss does not validate each mechanistic component: different combinations of \(aE\), \(q\), and \(r\) can yield the same hazard.

An explicit conjugate-state model is an alternative when contact duration and delayed execution matter. It can represent formation of recognized contacts, their dissociation, and subsequent target-cell death as distinct transitions, instead of compressing them into an effective per-encounter probability. The compact hazard model is simpler and may be appropriate when those intermediate dynamics are not the modeling objective or are not resolved by the available data. The choice is not simply a matter of detail: an explicit model introduces additional parameters that may be difficult to estimate without contact-level measurements.

A major failure mode is to interpret \(a\) as a killing rate. It describes encounters per effector–target opportunity under the chosen scaling; killing requires both recognition and successful execution. Another limitation is that the mass-action approximation does not represent finite contact time or a restriction on how many targets an effector can engage. If such effects matter, the assumption that available effectors sustain independent encounter events can fail, and the fitted \(a\) or \(r\) may absorb the mismatch. The model also does not resolve variation in antigen presentation among tumor cells; using a single \(q\) for a heterogeneous target population can conceal important differences.