Linking Protein Burden to Neuronal Injury and Loss

From Aggregate Burden to Neuronal Loss

Protein aggregation kinetics describe how aggregate burden changes over time. A separate modeling step is needed to link that burden to neuronal injury and, ultimately, to the number of surviving neurons. A high burden does not by itself specify how quickly cells are injured or lost: the relationship may depend on crossing a threshold, the duration of exposure, or the burden present at each moment.

Let \(B(t)\) denote the aggregate burden at time \(t\), measured on a defined scale. It may represent intracellular burden, extracellular burden, or a specified combination; the chosen measure and its units must be explicit. Let \(N(t)\) denote the number of viable neurons in the modeled population. The formulations below describe population-level trajectories. They do not, by themselves, identify which individual neurons are most vulnerable or establish that measured burden causes observed loss.

A model may connect \(B(t)\) directly to \(h(t)\), or introduce an injury state between burden and loss. This tutorial focuses on three useful burden-to-loss formulations: a threshold response, an accumulated-exposure response, and a smooth burden-dependent hazard. Their differences concern how past and present burden influence injury and loss—not simply which equation is easiest to fit.

Survival From a Specified Hazard

Before comparing burden-response models, it is useful to establish how a hazard determines the neuronal trajectory. Assume that \(h(t)\) is nonnegative, specified over the time interval of interest, and shared by the modeled neurons. The population equation \(dN/dt=-h(t)N(t)\) says that the instantaneous loss rate is the current population multiplied by the current hazard.

Separate variables and integrate from 0 to \(t\). This gives a direct relationship between cumulative hazard and surviving population:

Thus, the surviving fraction is \(N(t)/N(0)=e^{-H(t)}\), and the fraction lost by time \(t\) is \(1-e^{-H(t)}\). A hazard is not the same quantity as the fraction lost: a constant hazard of 0.1 per day does not mean that 10% of the original population is lost each day. The fraction remaining declines exponentially under a constant hazard.

Three Ways to Link Burden and Loss

A threshold formulation makes injury or loss increase only after burden crosses a specified level. One simple threshold hazard is \(h(t)=h_0+h_1\mathbf{1}\{B(t)>B_c\}\), where \(h_0\) is background hazard, \(h_1\) is the added hazard above the threshold, \(B_c\) is the threshold burden, and the indicator equals 1 when its condition holds and 0 otherwise. This is a sharp switch: burdens just below and just above \(B_c\) imply different hazards.

A cumulative-exposure formulation instead tracks burden accumulated over time. For a reference burden \(B_0\), define excess exposure \(A(t)=\int_0^t [B(s)-B_0]_+\,ds\), where \([x]_+=\max(x,0)\). A simple hazard link is \(h(t)=h_0+\gamma A(t)\), with \(\gamma\geq 0\). Here, past exposure continues to affect current hazard because it remains in \(A(t)\). This version has no repair or clearance of the injury-producing exposure state.

A smooth burden-dependent hazard responds to current burden without a hard threshold. One common saturating form is \(h(t)=h_0+h_{\max}B(t)^m/(K^m+B(t)^m)\), for nonnegative burden, \(K>0\), and \(m>0\). In this form, \(K\) is the burden producing half of the maximum added hazard, and \(m\) controls the steepness of the transition. Unlike cumulative exposure, the hazard depends on current burden, not on the entire history, unless a separate history-dependent state is added.

These equations are alternative hypotheses about the burden-to-injury link, not interchangeable parameterizations of one established mechanism. In particular, the threshold model above is a threshold in hazard, whereas a model that first produces a reversible injury state would need an additional state equation and its own recovery assumptions.

Worked Examples

Comparing Trajectories and Checking Model Behavior

The three formulations can produce different loss trajectories even when they use the same burden time course. A threshold model predicts a switch in hazard at \(B_c\), and its output may be sensitive to small changes in the chosen threshold. A cumulative-exposure model predicts that the hazard can keep rising while burden remains above \(B_0\), and can remain high after burden falls because accumulated exposure is retained. A smooth current-burden model predicts a continuous change in hazard as burden changes; saturation limits how much further the added hazard rises at high burden.

The hazard-to-population relationship provides a common basis for comparison: calculate \(H(t)\) under each model, then use \(N(t)=N(0)e^{-H(t)}\). A useful diagnostic is to plot burden, the implied hazard, cumulative hazard, and predicted \(N(t)/N(0)\) on compatible time axes. Check that hazards are nonnegative, units are consistent, and the predicted surviving fraction stays between 0 and 1 and never increases when there is no neuronal input. Abrupt jumps in the threshold-model hazard may be implausible if the biological process is expected to respond gradually; a very steep smooth curve may approximate such a switch, but does not make the threshold assumption automatically correct.

In data, compare predicted and observed neuronal-loss trajectories across time and, where available, across burden patterns that differ in peak, duration, or timing. If loss continues after burden falls, a purely current-burden model with only background hazard may miss that persistence; cumulative exposure or a damage state with slow recovery might represent it. Conversely, if two regimens with similar accumulated exposure but different peaks have different outcomes, a cumulative model that ignores peak burden may be inadequate. These patterns help assess model structure, but sparse measurements can make different structures difficult to distinguish.

Threshold and smooth hazard models are alternatives when the main question concerns how current burden relates to current loss. Cumulative exposure is a natural alternative when duration and history are important. A model with an explicit injury state and recovery is another possible structure, but it adds parameters and requires data that inform injury formation and resolution. No formulation is universally preferable: complexity should be justified by the intended use and by information in the data.

Limitations and Interpretation

These formulations simplify a heterogeneous biological system. A single burden measure may combine aggregates with different locations or properties, and a single population hazard may hide variation in cell susceptibility. The burden-response parameters can also be difficult to identify separately: for example, limited observations may not distinguish a sharp threshold from a steep smooth response, or a slowly changing burden effect from accumulated exposure. Measurement error or a mismatch between the modeled burden and the burden relevant to injury can further obscure the relationship.

A fitted relationship is not, on its own, proof that aggregate burden caused neuronal loss. The equations state model assumptions and generate predictions conditional on them. Their practical value is to make those assumptions explicit, quantify the resulting cell-loss trajectory, and reveal which additional observations—especially observations that vary burden magnitude, duration, and timing—would help discriminate among the alternatives.