Worked Model of a T-Cell Engager Across Dose and Antigen Density

From Engager Dose to Cell Killing

A T-cell engager must connect a T cell to a target cell for a productive interaction, but the concentration that occupies each binding arm is not necessarily the concentration that maximizes cell bridging. At high engager concentrations, the two cell types can become separately occupied by different engager molecules, leaving fewer available opportunities to form a bridge. A useful quantitative model should therefore distinguish arm occupancy, bridge abundance, and killing, rather than treating them as interchangeable measurements.

The model here is a deliberately reduced, hypothetical example. It compares two tumor-cell populations, one with low and one with high target-antigen density, as free engager concentration changes. Its equations are not a complete molecular binding model; their purpose is to make assumptions and predictions explicit.

A Competition-Adjusted Bridging Model

The occupancy equations are simple saturable binding expressions. To represent the loss of productive bridges when high free drug separately occupies the two cell types, the model includes a competition factor \(K_H/(K_H+C)\). This factor is a phenomenological reduction in bridge formation, not an additional measured arm occupancy. The resulting bridge equation is

For the worked setting, \(K_E=2\) ng/mL, \(K_{A,L}=10\) ng/mL, \(K_{A,H}=3\) ng/mL, and \(K_H=10\) ng/mL. The maximum bridge scales are \(B_{\max,L}=2350.53\) and \(B_{\max,H}=2366.112\) bridge units per target cell. These hypothetical scales represent population-specific capacities; they are not inferred from occupancy alone.

At low concentration, both occupancies are small, so bridge abundance initially increases with dose. At high concentration, the competition factor becomes small. In fact, as \(C\) becomes large, the two occupancy fractions approach one while \(K_H/(K_H+C)\) approaches zero. Thus \(B_j(C)\) eventually declines. This model can produce a bell-shaped bridge response without claiming that every engager has the same response curve.

Worked Example: Occupancy and Bridging at 5 ng/mL

Why the Dose Response Can Turn Down

The high-dose behavior follows directly from the bridge equation. Taking the logarithmic derivative with respect to \(C\) gives

At very low \(C\), the \(2/C\) term dominates, so bridge abundance rises with concentration. At very high \(C\), the four terms together behave approximately as \(2/C-3/C=-1/C\), which is negative. The bridge abundance must therefore turn downward somewhere between those regions. For the low-density parameters, setting this derivative to zero yields \(C^2-10C-40=0\); the positive solution is \(C=(10+\sqrt{260})/2=13.06\) ng/mL. This is the model's low-density bridge-abundance peak, not a universal optimal clinical dose.

From Bridge Abundance to Selective Killing

The killing link used above is a constant-hazard model. If each cell has a constant instantaneous killing hazard \(k_j\), then the probability of remaining alive through time \(t\) is \(S_j(t)=\exp(-k_jt)\), and the fraction killed is \(1-S_j(t)\). This follows from the survival equation \(dS_j/dt=-k_jS_j\) with \(S_j(0)=1\): separating variables and integrating gives \(\log S_j(t)=-k_jt\), hence the exponential expression. The units are consistent because \(k_j\) is per hour and \(t\) is in hours.

In this example, \(k_j=\gamma B_j\) assumes killing hazard is directly proportional to bridge abundance. That is a simplifying assumption, not a general law. It is most defensible over a range where bridge abundance is a useful proxy for productive contacts and where effector-cell availability is not limiting. If killing saturates, an alternative could be \(k_j=k_{\max}B_j/(B_{50}+B_j)\), with \(B_{50}\) the bridge abundance giving half of the maximum hazard. Such a model needs additional parameters and data.

Diagnostics, Alternatives, and Limitations

A model is useful only if its assumptions remain credible for the question being asked. Check whether free concentration is approximately maintained, whether target and effector populations are sufficiently well mixed, and whether measured bridge abundance or killing supports the assumed dose dependence. If observed killing continues to rise at high concentrations while this model predicts a decline, possible explanations include an inappropriate competition factor, changing exposure, or biology absent from the reduced model. A mismatch is a reason to revisit assumptions, not to relabel the prediction as an observation.

An occupancy-only product, \(\theta_E\theta_{A,j}\), is a simpler alternative. It is easier to interpret but approaches one at high concentration and cannot represent the high-dose decline in bridges. A more detailed binding and cell-interaction model could track free drug, singly bound species, bridges, receptor turnover, and cell encounters explicitly. Such a model can represent more mechanisms, but it requires more data and identifiable parameters. The present reduced model is a transparent compromise; its competition factor and calibrated bridge scales should not be mistaken for uniquely established molecular quantities.

Other limitations matter for interpretation. The two populations are each treated as uniform, although actual tumors may contain a distribution of antigen densities. The model does not represent spatial access, effector-to-target ratios, serial killing, cytokine effects, or time-varying exposure. The assumed linear conversion from bridges to killing may fail when effector cells are saturated or when signaling has thresholds. Consequently, the predicted selectivity is conditional on the model structure and parameterization.

Common Pitfall

• Wrong reasoning: “At 20 ng/mL, both binding arms are more occupied than at 5 ng/mL, so there must be more bridges and more killing.”
• Correction: Arm occupancy and bridge abundance are distinct. In this model the competition factor falls from \(10/15=0.6667\) at 5 ng/mL to \(10/30=0.3333\) at 20 ng/mL. The resulting high-density bridge prediction falls from 704.2 to 623.5 units per cell, and the predicted 24-hour killing falls from 81.55% to about 77.61%.