Translating Intracellular Payload Exposure to Cell Killing

From Intracellular Payload to Cell Killing

An intracellular payload concentration is not itself a measure of cell death. Payload must produce a pharmacological effect, and that effect may take time to develop. A concentration measured now can therefore be associated with killing later; two profiles with the same total concentration-time area can also produce different killing over a specified observation period.

The earlier material on intracellular payload release provides the conceptual basis for an intracellular concentration time course. Here, take that time course as the input and translate it into a delayed killing rate and a viable-cell trajectory. The focus is the link from exposure to response, not the upstream mechanisms that generate payload.

Definition: Let \(C_i(t)\) be intracellular payload concentration, in nM, and let \(C_e(t)\) be an effect-site concentration, also in nM. \(C_e(t)\) is a model state that represents delayed pharmacological drive; it need not be a separately measurable physical compartment. Let \(N(t)\) be the number of viable cells. A delayed exposure-response model specifies how \(C_i(t)\) changes \(C_e(t)\), how \(C_e(t)\) determines the killing rate, and how killing changes \(N(t)\).

A Minimal Delayed Exposure-Response Model

A standard effect-compartment link lets the effect-site state approach the intracellular concentration at rate \(k_{e0}\):

\frac{dC_e}{dt}=k_{e0}(C_i-C_e).

Here \(k_{e0}>0\) has units of inverse time. A small \(k_{e0}\) means a slower response to changing intracellular concentration. This equation assumes a linear, first-order approach to equilibrium, with no separate effect-site input or loss mechanism beyond the equilibration represented by \(k_{e0}\). If \(C_i\) is constant over an interval of length \(\Delta t\), the exact update is:

C_e(t+\Delta t)=C_i+\{C_e(t)-C_i\}e^{-k_{e0}\Delta t}.

A common concentration-response function for the payload-induced killing rate is the Hill Emax model:

k_{\mathrm{kill}}(t)=\frac{k_{\max}C_e(t)^h}{EC_{50}^h+C_e(t)^h}.

The parameter \(k_{\max}\) is the maximum killing rate, in day\(^{-1}\) if time is in days; \(EC_{50}\) is the effect-site concentration producing half of \(k_{\max}\); and \(h>0\) controls the steepness of the concentration-response curve. This model assumes killing rate is a nonnegative, instantaneous function of \(C_e\), with no additional dependence on cell state or time. It does not imply that \(EC_{50}\) is the concentration producing 50% cell death.

If viable cells also proliferate at a constant rate \(r\), then:

\frac{dN}{dt}=\{r-k_{\mathrm{kill}}(t)\}N.

For \(N(0)>0\), integration gives the log-ratio and the cumulative killing hazard \(H(T)\):

\log\!\left(\frac{N(T)}{N(0)}\right)=rT-H(T), \qquad H(T)=\int_0^T k_{\mathrm{kill}}(t)\,dt.

When \(r=0\), the surviving fraction is \(N(T)/N(0)=e^{-H(T)}\), and the fraction killed is \(1-e^{-H(T)}\). With proliferation, a change in cell count reflects both growth and killing; it is not the killing fraction alone.

Why the Calculation Uses Cumulative Hazard

The cell-count equation is separable: dividing both sides by \(N(t)\) gives \(dN/N=\{r-k_{\mathrm{kill}}(t)\}dt\). Integrating from 0 to \(T\) yields the log-ratio above. This shows why an instantaneous killing rate cannot by itself describe the final response: the rate must be accumulated over the observation interval. It also makes the role of the observation window explicit. Killing that occurs after \(T\) does not contribute to \(H(T)\).

In practical calculations, \(C_e(t)\) and \(k_{\mathrm{kill}}(t)\) are often evaluated on a time grid. The effect-compartment update can be exact for piecewise-constant \(C_i\), but the hazard integral still needs numerical integration. The examples below use a right-endpoint approximation: each interval's killing rate is approximated by its value at the interval end, multiplied by the interval length. This is transparent but can be inaccurate when rates change rapidly or intervals are long.

Worked Example: Constant Effect-Site Concentration

Suppose \(C_e\) is maintained at 2 nM for 3 days after the effect-site delay has passed. Let \(k_{\max}=0.40\) day\(^{-1}\), \(EC_{50}=2\) nM, \(h=1\), and assume no proliferation. Calculate the predicted fraction killed.

At 2 nM, the killing rate is:

k_{\mathrm{kill}}=\frac{0.40(2)}{2+2}=0.20\ \text{day}^{-1}.

The rate is constant, so \(H(3)=0.20\times3=0.60\). The surviving fraction and fraction killed are:

\frac{N(3)}{N(0)}=e^{-0.60}=0.5488, \qquad 1-e^{-0.60}=0.4512.

The model predicts about 45.1% killing over 3 days. This is a model-based fraction over that interval, not the instantaneous effect at 2 nM.

Matched Total Exposure, Different Timing

Total intracellular exposure over an interval is often summarized by the area under the concentration-time curve, \(\int_0^T C_i(t)\,dt\). Equal areas do not guarantee equal \(H(T)\). The effect-site state weights concentration according to its timing, the concentration-response curve may be nonlinear, and the observation window may end before delayed effects from late exposure have occurred.

Worked Example: Equal Payload AUC, Early Versus Late Release

Consider two hypothetical intracellular profiles over 3 days, each expressed as a constant concentration within each 1-day interval. Profile A is \(C_i=(6,0,0)\) nM; profile B is \(C_i=(0,0,6)\) nM. Both have an AUC of \(6\) nM·day. Let \(C_e(0)=0\), \(k_{e0}=\ln(2)\) day\(^{-1}\), \(k_{\max}=0.40\) day\(^{-1}\), \(EC_{50}=1\) nM, \(h=1\), and \(r=0\).

For each interval, \(e^{-k_{e0}\Delta t}=0.5\), so the exact effect-site update is \(C_e(t+1)=C_i+0.5\{C_e(t)-C_i\}\). For Profile A, the effect-site values at the interval ends are 3, 1.5, and 0.75 nM. For Profile B, they are 0, 0, and 3 nM.

Using the killing rate at each interval end as a right-endpoint approximation, Profile A has rates \(0.40(3)/(1+3)=0.300\), \(0.40(1.5)/(1+1.5)=0.240\), and \(0.40(0.75)/(1+0.75)=0.1714\) day\(^{-1}\). Thus:

H_A(3)\approx(0.300+0.240+0.1714)(1)=0.7114, \qquad \frac{N_A(3)}{N_A(0)}\approx e^{-0.7114}=0.4909.

For Profile B, the corresponding rates are 0, 0, and \(0.40(3)/(1+3)=0.300\) day\(^{-1}\). Therefore:

H_B(3)\approx0.300, \qquad \frac{N_B(3)}{N_B(0)}\approx e^{-0.300}=0.7408.

The model predicts about 50.9% killing for early release and 25.9% for late release by day 3, despite matched payload AUC. The late profile has less time to develop killing before the endpoint. This is a finite-window comparison: the calculation does not assert that responses at every later time will retain the same difference. The right-endpoint approximation also means these figures are illustrative numerical predictions, not exact integrals of the continuous killing rate.

Separating Killing From Net Cell Growth

Cell counts can rise even when a payload kills cells, if proliferation is faster than killing. Conversely, a declining count does not identify the killing rate without accounting for baseline growth. The next calculation shows how the model uses a growth assumption to interpret a measured count ratio.

Worked Example: Inferring a Killing-Rate Scale

In a hypothetical experiment, effect-site concentration is held at 3 nM for 4 days, after equilibration. The viable-cell count at day 4 is 60% of its initial value. Assume \(r=0.050\) day\(^{-1}\), \(EC_{50}=3\) nM, \(h=1\), and the constant-\(C_e\) model. Estimate \(k_{\max}\).

At 3 nM, the Hill fraction is \(3/(3+3)=0.5\), so \(k_{\mathrm{kill}}=0.5k_{\max}\). The cell-count equation gives:

\log(0.60)=\{0.050-0.5k_{\max}\}(4).

Since \(\log(0.60)=-0.5108\), division by 4 gives \(-0.1277=0.050-0.5k_{\max}\). Rearranging:

k_{\max}=\frac{0.050+0.1277}{0.5}=0.3554\ \text{day}^{-1}.

Under the stated assumptions, the estimated maximum killing rate is 0.3554 day\(^{-1}\). This is not an assumption-free estimate: it depends on the specified growth rate, constant effect-site concentration, and concentration-response form. Different growth or exposure assumptions could give a different estimate from the same observed cell-count ratio.

Assumptions, Diagnostics, and Alternatives

Before interpreting a prediction, check that the model represents the experiment and that its inputs are adequate. Useful questions include:

Is the input concentration credible? Intracellular total payload and pharmacologically available payload need not be interchangeable. Use the concentration measure that is relevant to the intended exposure-response model.
Is the delay plausible? If observed killing begins before the modeled \(C_e\) can rise, or persists longer than the model predicts, the effect-compartment time scale may be inadequate.
Does the concentration-response shape fit? Systematic residual patterns across concentrations, or implausible estimates of \(EC_{50}\), \(h\), or \(k_{\max}\), can indicate that the Hill model is insufficient or that the data do not identify its parameters.
Are cell counts being compared at appropriate times? Sparse measurements can miss an early response or delayed killing. Check predictions against the full time course, not only a single endpoint.
Are cells homogeneous enough for the model? A single \(N(t)\) and one set of killing parameters describe an average population. A distribution of sensitivities or exposure across cells can make the average response depart from the response predicted for a representative cell.

The effect-compartment model is a compact way to represent delay. An alternative is a direct, no-delay model in which \(C_e=C_i\); it is simpler but cannot represent a lag between concentration and response. A more mechanistic alternative adds explicit intracellular damage or repair states. Such models can represent persistent lesions or recovery, but require additional assumptions and data to estimate their parameters. These are alternatives, not automatic improvements: a more elaborate model is useful only when its added states are supported by observations.

Important limitations remain even when the equations fit. The model assumes a chosen functional form for killing, a common response across modeled cells, and specified proliferation behavior. It does not by itself distinguish cell death from loss of viability signal, establish why two payload profiles differ, or validate extrapolation outside the measured concentration and time ranges. In particular, the finite observation period can be decisive when release occurs late.

Common Pitfall

Wrong reasoning: “The two treatments have the same intracellular AUC, so they must kill the same fraction of cells.” This conclusion treats concentration area as though it were the killing hazard. It ignores effect-site delay, the nonlinear concentration-response function, and when the response is measured. Correction: calculate or simulate \(C_e(t)\), convert it to \(k_{\mathrm{kill}}(t)\), and integrate that rate over the specified observation interval. Compare cell survival using the resulting cumulative hazard and the same growth assumptions.

Key takeaway: Intracellular payload concentration becomes a prediction of cell killing only after specifying the delay, concentration-response relationship, cell-growth behavior, and observation window. Matched payload AUC alone does not determine matched cell survival.

Check Your Understanding

Answer each question, then check the concise solutions below.

If \(k_{e0}\) decreases while the intracellular concentration profile is unchanged, what happens to the speed with which \(C_e\) follows \(C_i\)?
With no proliferation, a constant killing rate of 0.10 day\(^{-1}\) acts for 5 days. What fraction of cells is predicted to survive?
Why can equal intracellular AUCs lead to different killing by a fixed endpoint?
If \(N(T)/N(0)=0.80\), \(r=0\), and \(T=2\) days, calculate the cumulative killing hazard.

Solutions. (1) \(C_e\) responds more slowly because its rate of approach is proportional to the smaller \(k_{e0}\). (2) \(H=0.10(5)=0.50\), so survival is \(e^{-0.50}=0.6065\), or 60.7%. (3) The delay, nonlinear rate function, and finite observation window can weight the timing of exposure differently. (4) From \(N(T)/N(0)=e^{-H(T)}\), \(H(2)=-\log(0.80)=0.2231\).