PD-1 Pathway Suppression and Blockade

From PD-1 Engagement to T-Cell Function

Activated T cells integrate activating inputs with inhibitory signals. Engagement of the inhibitory receptor PD-1 by either PD-L1 or PD-L2 can attenuate T-cell signaling and reduce functional outputs such as cytokine production or proliferation. A quantitative systems pharmacology model can represent this pathway in two linked steps: ligand engagement generates a suppressive signal, and that signal reduces the output associated with T-cell activation.

This representation is deliberately narrower than a full immune-response model. It concerns suppression of an already activated T-cell functional response; it does not model how T cells are primed, how chronic antigen exposure changes cell states, or how T cells kill tumor cells. Those processes require additional state variables and mechanisms.

The equations are phenomenological at the level of cellular function: they do not resolve every intracellular signaling step. Their purpose is to make explicit how ligand abundance, binding affinity, and a blocking drug can influence a defined T-cell output.

Competitive Binding and the Suppressive Signal

Consider a single, well-mixed receptor pool with PD-L1, PD-L2, and a blocking drug that compete for mutually exclusive binding to PD-1. Let \(L_1\) and \(L_2\) denote the free concentrations of PD-L1 and PD-L2, and let \(B\) denote the free concentration of a PD-1-blocking drug. Let \(K_{d1}\), \(K_{d2}\), and \(K_{dB}\) be their respective dissociation constants. Define the dimensionless ratios \(x_1=L_1/K_{d1}\), \(x_2=L_2/K_{d2}\), and \(x_B=B/K_{dB}\).

There are four receptor states in this model: unbound, bound by PD-L1, bound by PD-L2, or bound by blocker. Their relative weights are \(1\), \(x_1\), \(x_2\), and \(x_B\). Normalizing these weights gives the fractions \(f_1\), \(f_2\), and \(f_B\) of receptors bound to PD-L1, PD-L2, and blocker:

The unbound fraction is \(f_0=1/D\), so \(f_0+f_1+f_2+f_B=1\). The total ligand-engaged fraction is \(q=f_1+f_2\). To allow the ligands to produce different strengths of inhibitory input, define \(S=\epsilon_1f_1+\epsilon_2f_2\), where \(\epsilon_1\) and \(\epsilon_2\) are nonnegative dimensionless signal weights. If the model assumes equal signal strength per occupied receptor, set both weights to 1, giving \(S=q\).

To connect this signal to an activated T-cell output, let \(a\) be a dimensionless activation drive summarizing activating inputs, and let \(Y_{\max}\) be the maximum output on the chosen measurement scale. A simple saturable activation term, \(a/(1+a)\), is multiplied by an inhibitory factor:

Here \(K_S>0\) is the suppressive-signal scale at which the inhibitory factor is one-half. The output \(Y\) has the same units as \(Y_{\max}\), such as a specified cytokine concentration or a proliferation readout. The model predicts less output as \(S\) increases, while preserving the activating drive as a distinct input. It does not assume blockade will restore function beyond the level permitted by \(a\), or remove inhibition from pathways other than PD-1.

Why Blockade Reduces the Modeled Inhibitory Input

The competitive-binding model yields a direct monotonicity result. Write \(P=x_1+x_2\). Then \(q=P/(1+P+x_B)\). For fixed ligand ratios, increasing blocker concentration increases \(x_B\) and reduces the ligand-engaged fraction:

For unequal ligand weights, let \(W=\epsilon_1x_1+\epsilon_2x_2\). The suppressive signal is \(S=W/(1+P+x_B)\), so \(\partial S/\partial x_B=-W/(1+P+x_B)^2\leq 0\). Finally, the functional output increases as the suppressive signal falls, provided activation drive and the other model inputs are held fixed:

Together, these results justify the model’s directional prediction: under its assumptions, increasing PD-1 blocker concentration cannot increase ligand occupancy, and reducing the modeled PD-1 suppressive signal cannot decrease the predicted activated-cell output. The prediction is conditional on fixed ligand levels, fixed activation drive, and the stated binding and response functions.

Assumptions, Diagnostics, and Alternative Formulations

The competitive equilibrium equations are useful only if their assumptions are a reasonable approximation for the question and data at hand. At minimum, check whether ligand and drug concentrations are available on a consistent scale, whether they can reasonably be treated as free concentrations, and whether receptor binding is fast enough relative to the timescale of interest to approximate equilibrium. The model also treats the receptor pool as homogeneous and the ligands as mutually exclusive competitors for a shared site. If receptor expression or local ligand availability varies substantially across cells or tumor regions, a single average fraction can hide important heterogeneity.

Useful diagnostics include checking that all calculated fractions lie between 0 and 1 and sum to 1, checking that \(S\) is nonnegative, and confirming that predicted output moves in the expected direction when blocker concentration changes while other inputs are fixed. When fitting data, compare model predictions with measurements across a range of ligand and drug conditions. Systematic underprediction or overprediction at high ligand levels, for example, may indicate that the assumed binding, signal weighting, or response function is too simple. A good numerical fit alone does not establish that a particular mechanism is uniquely identified by the data.

There are several reasonable alternatives, each answering a different modeling need:

The simple equilibrium model also has limits. It does not include binding kinetics, receptor internalization, changes in receptor expression, spatial gradients, cell-to-cell variability, or feedback that changes ligand abundance. It summarizes inhibition as one signal rather than modeling intracellular signaling in detail. An anti-PD-L1 intervention can be represented by reducing available PD-L1 in this framework, but it does not directly remove PD-L2; a PD-1 blocker competes at the receptor in the formulation above. Neither representation guarantees that observed clinical response will follow the predicted cellular output, because other biological processes and exposure patterns are outside this model.

Common Pitfall

A common mistake is to treat the blocker-bound fraction as the fraction of T-cell function restored. For example, in the second worked example \(f_B=0.4444\), but the model predicts an output of \(0.3158Y_{\max}\), not a 44.44% restoration. The blocker fraction describes receptor binding under the occupancy assumptions; the functional output also depends on remaining ligand engagement, the signal scale \(K_S\), and activation drive \(a\). Keep these quantities distinct, and state which one is being predicted or measured.