Representing Tissue Geometry and Microenvironments

Representing Geometry at the Scale the Question Requires

A spatial model needs a domain: the set of locations at which its state variables are defined. That domain is a deliberate representation of tissue, not a claim that every anatomical feature has been reconstructed. Its dimensionality and compartments should be chosen to capture the spatial differences that could change the model’s answer. A model of penetration across a thin tissue layer may need only one spatial coordinate. A model asking whether a signal reaches cells around a localized source may need two or three.

The earlier material on choosing spatial scales and state variables established that spatial resolution should follow the mechanism and the question. Here the focus is how to turn that choice into a domain, how to describe regions and their interfaces, and how much spatial variation to assign to parameters. The key constraint is evidential: adding a detailed boundary or parameter map is not an improvement if its location or value is unsupported.

A domain may be a line segment, a planar region, or a volume. Those are modeling choices, not statements that tissue is literally one- or two-dimensional. A one-dimensional domain represents variation along one coordinate while treating variation in other directions as negligible or averaged. A two-dimensional domain represents variation across a plane, often with an implicit assumption that out-of-plane variation is negligible. A three-dimensional domain can represent variation in all spatial directions, but it requires enough information to define that geometry and enough computational and measurement support to use it.

Compartments also need not correspond one-to-one with named anatomical structures. A compartment can instead mean a region assigned a distinct transport property, source term, or state-variable definition. Its biological interpretation should be no more specific than the evidence permits. Likewise, a spatially varying parameter can represent a known gradient or a deliberately simplified pattern; it should not imply a finely mapped tissue property when only an average is known.

Choosing Dimensionality and Defining Regions

Begin with the spatial contrasts that matter to the prediction. If the output depends mainly on distance from an exposed boundary, a one-dimensional slab may capture the relevant contrast. If it depends on lateral separation between a source and nearby cells, a planar domain may be needed. If pathways around an obstacle or differences between surface and interior matter, three dimensions may be necessary. Dimensionality is therefore determined by the question and the evidence, not by a general preference for the most detailed model.

A reduced-dimensional model is most defensible when omitted directions are approximately uniform, symmetric, or represented by a justified average. In a one-dimensional slab, for example, concentration is assumed not to vary appreciably across the slab’s width and height. A three-dimensional model is not automatically more accurate: uncertain geometry, boundary placement, and parameter fields can make a detailed representation less reliable than a transparent reduced model.

To define compartments, specify their extent and the rule that distinguishes them. A partition of \(\Omega\) into regions \(\Omega_i\) should avoid accidental gaps or overlaps unless those are intended. A region may have a constant parameter, or a parameter field may vary within it. At an interface, state whether the model permits free passage, a finite resistance, a prescribed flux, or no passage. These choices are part of the model specification, not consequences of drawing a boundary.

Interfaces Encode Exchange, Not Just Contact

When two modeled regions meet, their interface conditions determine how state variables and fluxes behave there. For a diffusing concentration \(c\), the flux vector is \(\mathbf{J}=-D\nabla c\). With a unit normal \(\mathbf{n}\) directed from region \(i\) toward region \(j\), the normal flux is \(\mathbf{J}\cdot\mathbf{n}\). A common idealization is perfect contact: concentration is continuous across the interface, and normal flux is conserved. This represents no additional resistance at the boundary.

A different idealization allows a concentration jump while relating flux to that jump through an interface permeability \(P\). Under a simple linear exchange law, flux from \(i\) to \(j\) is \(P(c_i-c_j)\), where \(c_i\) and \(c_j\) are the concentrations on the two sides. Here \(P\) has units of length per time when concentration is amount per volume and flux is amount per area per time. A barrier with no exchange is a limiting case with zero flux. The appropriate condition depends on the mechanism and available evidence; the mere presence of a compartment boundary does not establish a barrier.

A useful justification for compartment exchange equations follows from conservation. If the interface area is \(A\), the exchange flux is uniform over that area, and the flux law is \(P(c_i-c_j)\), then the amount transferred per time is flux times area. Dividing by a compartment volume converts amount per time into concentration change per time:

The sign is positive for compartment \(i\) when \(c_j>c_i\). The factor \(A/V_i\) has units of inverse length, so multiplying it by \(P\) gives inverse time. The same interface contributes an equal and opposite amount-per-time term to compartment \(j\), preserving total amount in the absence of other sources, sinks, or boundary losses. This derivation assumes a well-mixed concentration within each compartment and a uniform interface flux; if concentrations vary strongly within a compartment, a continuous spatial description may be needed instead.

Assigning Spatially Varying Parameters

A parameter need not be constant throughout \(\Omega\), but its representation should match what is known. Three common levels are a single value throughout the domain, piecewise-constant values assigned by compartment, and a continuous field such as a smooth gradient. A constant parameter is the simplest assumption. Piecewise values are useful when evidence supports a distinction between regions but not variation within them. A continuous field is appropriate when the model or data supports a spatial trend and its scale.

A discontinuity at an interface is not merely a plotting choice: it can change transport, local reaction rates, or the predicted distribution. State whether a parameter is intended to jump, remain continuous, or change smoothly. If the evidence supports only an average, assigning a detailed spatial map creates apparent precision without information. It is often better to represent uncertain heterogeneity by a small number of interpretable parameters and test how predictions respond to them.

Diagnostics, Alternatives, and Limitations

Geometry should be checked against both the biology and the mathematical consequences of the representation. Confirm that region extents and units are consistent, that interface normals and flux signs are defined consistently, and that the proposed compartments have a clear role in the model. For a reduced-dimensional domain, assess whether omitted-direction variation is plausible given the question and observations. For a compartment model, check whether its well-mixed assumption is credible relative to the transport and reaction timescales. Sensitivity to region boundaries, interface permeability, and parameter contrast can reveal whether an apparently precise conclusion depends on weakly supported choices.

The main alternatives serve different purposes. A continuous domain represents gradients and local spatial variation directly. A compartment model averages within regions and makes exchange between them explicit. A hybrid representation can use continuous fields in some regions and compartment-level exchange elsewhere, but introduces additional coupling assumptions. The introductory material on when spatial compartments approximate continuous tissue provides the broader interpretation of this trade-off; here the practical point is to choose the simplest representation that preserves the contrasts relevant to the question.

These representations have limits. A one-dimensional model cannot predict lateral localization. A two-dimensional model cannot resolve out-of-plane pathways unless those effects are parameterized. Compartments can hide within-region gradients, and a finely varying parameter field can be unidentifiable from sparse measurements. Geometry and parameters may also be confounded: changing a boundary location can resemble changing a transport rate. Numerical refinement cannot correct a biologically unsupported domain or interface condition.