Why Spatial Scale and State Definition Come First
A spatial QSP model is not defined by adding a location label to an ordinary time-course model. It requires decisions about where each state is defined, over what physical scale it represents, and which biological quantity it tracks. These choices determine whether a model can represent differences between cells, gradients across tissue, or both.
The earlier material on when spatial heterogeneity changes model predictions motivates these choices: averaging can conceal biologically important variation. The task here is to make the spatial representation precise before specifying the mechanisms that change states over time. A model intended to predict average tissue exposure may not need the same representation as one intended to predict which individual cells cross a signaling threshold.
Coordinates and Representative Length Scales
A coordinate system should correspond to the physical question, not merely to a convenient mathematical form. For a thin tissue slice, one coordinate \(x\) measured from the exposed surface may be sufficient if the important change is with depth. If variation in two in-plane directions matters, coordinates \((x,y)\) may be needed. The model specification should state the origin, axis directions, and units; otherwise, a reported location such as \(x=0.5\) is ambiguous.
Several distinct length scales can matter in one tissue: a cell diameter \(a\), the spacing between cells, the distance over which cell properties remain correlated, and the length \(L\) over which an extracellular signal changes appreciably. These scales need not be equal. A continuum description of local averages is well motivated when there is a range of averaging widths \(\ell\) that are large compared with individual cells but small compared with the tissue feature of interest:
Here, \(\ell\) is the spatial support of a local average, not automatically the size of a numerical grid element. If no such separation exists, a smooth continuum field may blur cell-to-cell differences or a sharp tissue transition. A cell-level representation, or a representation that keeps distinct tissue regions, may then be more appropriate.
Worked Example: Is There a Clear Scale for a Continuum Average?
Suppose a hypothetical tissue is \(1.2\) mm thick. Its cells are about \(20\) μm in diameter, while measured extracellular drug concentration changes substantially over a characteristic depth of about \(0.30\) mm. Assess whether a local average over \(\ell=60\) μm has a clear scale separation.
Convert the lengths to the same units: \(20\) μm is the cell diameter, \(60\) μm is the averaging width, and \(0.30\) mm is \(300\) μm. Thus,
The averaging width spans only about three cell diameters, while the gradient length is only about five averaging widths. Neither ratio is especially large, so the condition \(a \ll \ell \ll L\) is not strongly supported. A continuum field may still be a useful approximation, but it should not be described as averaging over many cells while remaining very local relative to the gradient. If predictions depend on individual-cell exposure, represent cells explicitly; if the question concerns broad depth trends, compare the continuum approximation with a coarser region-based description.
Choosing Local State Variables
A state variable should identify both the biological quantity and its spatial meaning. “Drug concentration” is incomplete if it could mean extracellular concentration, intracellular concentration, or an average over tissue volume. Similarly, “receptor occupancy” could mean the occupancy fraction of one cell, the mean across cells near a location, or the fraction of all receptors in a tissue region that are bound. Those quantities can differ even when they use the same label.
For each state, specify its units, the compartment or population to which it refers, and the spatial support over which it is defined. Concentrations might be expressed per extracellular volume; cell densities as cells per tissue volume; and cellular signaling states as amounts or fractions per cell. Fractions are dimensionless, but their population being averaged still needs definition. If cells differ in receptor abundance, the mean cell occupancy and the fraction of all receptors occupied are not generally equal.
Consistency of scale is essential when connecting states. An amount per cell cannot be added directly to a concentration per tissue volume. To convert, the model needs the relevant cell density and volume fractions. This is a bookkeeping requirement, not a mechanistic assumption.
Worked Example: Converting a Per-Cell State to a Tissue Concentration
In a hypothetical tissue region, there are \(1.0\times10^6\) cells per mL of tissue. Each cell has an average of \(2.0\) fmol of receptor-bound ligand. The extracellular volume fraction is \(0.20\). Calculate the bound ligand per tissue volume and the corresponding concentration per extracellular volume.
First multiply the amount per cell by the number of cells per tissue volume:
One mL of tissue contains \(0.20\) mL of extracellular volume. Therefore, the bound amount expressed per extracellular volume is
The same bound ligand amount is \(2.0\) nmol per mL of tissue, or \(10\) μM when expressed per mL of extracellular volume. These are different volume bases, not conflicting answers. If a model uses extracellular concentration, it must retain the extracellular volume fraction when relating that state to the tissue-level amount.
Justifying a Local Average
A local average is useful when it suppresses fine-scale variation without erasing the larger spatial pattern of interest. A simple calculation shows why smoothness and scale separation matter. Consider a one-dimensional state \(C(x)\) averaged uniformly over a symmetric interval of width \(\ell\) centered at \(x\). Assume \(C\) is sufficiently smooth on that interval. A Taylor expansion gives \(C(x+u)\) as its value at the center plus a linear term and a curvature term. The linear term averages to zero because the interval is symmetric.
The leading difference between the local average and the central value is therefore controlled by curvature and by the square of the averaging width. This approximation is informative only when higher-order changes across the interval are small; it does not justify averaging across a sharp boundary or a narrow transition.
Worked Example: Checking the Error From a Local Average
Suppose, hypothetically, that drug concentration near a location follows \(C(x)=10+4x+2x^2\) μM, with \(x\) measured in mm. Estimate the average across a symmetric interval of width \(0.40\) mm centered at \(x=0.50\) mm, and compare it with the center value.
At the center, \(C(0.50)=10+4(0.50)+2(0.50)^2=12.5\) μM. The second derivative is constant: \(C''(x)=4\) μM/mm². Using the symmetric-average approximation,
The average differs from the center value by about \(0.0267\) μM, or \(0.21\%\) of \(12.5\) μM. For this smooth profile and chosen width, the local average is close to the center value. That conclusion would not automatically hold near a steep interface, where curvature can be large or the Taylor approximation can fail.
Matching the Representation to the Question
Three common representations emphasize different information. A continuum field assigns local average states throughout tissue and is suited to questions about spatial gradients when local averaging is defensible. A cell-level model assigns states to individual cells and is suited to cell-to-cell variability, discrete interactions, or thresholds that depend on individual cells. A spatial-compartment model assigns states to a finite set of regions and is suited to contrasts among zones when within-zone variation is not central.
| Representation | What its local state means | Main scale-related risk |
|---|---|---|
| Continuum field | A smoothly varying local quantity, often an average over many cells | Cell-level variability or sharp transitions may be smoothed away |
| Cell-level | A quantity for a particular cell at its position | Using a single-cell state may omit tissue-level variation unless cell positions and the relevant domain are represented |
| Spatial compartments | A mean or total for a defined tissue region | Within-region gradients may be hidden by the regional average |
The choice should follow the model output of interest. If the decision depends on the fraction of cells exceeding a signaling threshold, a regional mean may not be enough: two regions with the same mean can have different distributions across cells. If the decision depends only on an organ-level average, resolving every cell may add complexity without improving the target prediction.
Diagnostics, Failure Modes, and Limits
Before interpreting predictions, check that the coordinate frame, length units, state units, and averaging support are explicit. Where data permit, compare variation at the chosen scale with variation at a finer or coarser scale. A profile that changes abruptly within one proposed averaging region is a warning that a smooth local-state interpretation may be misleading. Likewise, unexplained dependence of predictions on the chosen region width can indicate that the model’s representation is not robust to its scale choice.
- Scale mismatch: A model claims to represent cell-to-cell variation but uses only one averaged state per large tissue region. Its outputs cannot recover individual-cell differences that the state definition has removed.
- Ambiguous concentration basis: An extracellular concentration is combined with an amount per tissue volume without accounting for the extracellular fraction. The resulting balance is dimensionally or biologically inconsistent.
- Overinterpretation of averages: A local mean is treated as the state of every cell in that region. This is not justified when cells vary substantially or responses are nonlinear.
- Unstated coordinates: A spatial profile is reported without its origin, direction, or units. Its locations cannot be mapped unambiguously to the tissue.
No single representation is universally best. The choice is limited by the biological question, the availability and resolution of spatial measurements, and the costs of added detail. A more detailed state definition does not guarantee better predictions: parameters for cell-to-cell variation or local quantities may not be identifiable from aggregate data. State explicitly which variation is represented, which is averaged out, and what observations could evaluate that decision.
Check Your Understanding
Answer each question, then compare with the solutions below.
- A tissue has cells about \(15\) μm across and a drug gradient length of \(45\) μm. Does an averaging width of \(30\) μm provide strong scale separation? Explain using ratios.
- Why is “mean receptor occupancy” incomplete as a state-variable definition when cells differ in receptor abundance?
- A model tracks ligand amount per cell but reports concentration per extracellular volume. What additional quantities are needed for a consistent conversion?
- For a smooth spatial profile averaged over a symmetric interval, why does the first derivative not contribute to the leading average-versus-center difference?
Solutions. (1) No. The width is \(30/15=2\) cell diameters and \(45/30=1.5\) times smaller than the gradient length, so neither \(a\ll\ell\) nor \(\ell\ll L\) is convincing. (2) The mean occupancy across cells weights each cell equally, whereas the fraction of all receptors occupied gives more weight to cells with more receptors; the intended population summary must be stated. (3) The cell density per tissue volume and the extracellular volume fraction are needed, along with consistent units. (4) Over a symmetric interval, positive and negative offsets have equal weight, so the linear Taylor terms cancel in the average; curvature is the leading contribution.