Choosing Spatial Resolution for Antibody Tissue Models

Why Spatial Resolution Is a Modeling Choice

A tissue compartment in a PBPK model may represent an entire organ or tissue region with a single antibody concentration. This is useful when transport within that region is fast enough to keep concentrations nearly uniform. If antibody enters slowly, moves only a short distance before binding or being removed, or encounters targets concentrated in particular regions, a single average can conceal biologically important gradients.

The choice is not simply between a “simple” model and a “detailed” one. Spatial resolution adds parameters and computational cost, and it is only useful when spatial differences affect the predictions of interest and can be supported by evidence. The earlier material on binding-site barriers in antibody tissue penetration establishes why local binding and uptake can restrict penetration; the present question is when that spatial process must be represented explicitly.

A well-mixed model does not claim that the real tissue is perfectly uniform. It claims that any nonuniformity is small enough, or sufficiently averaged, that it does not materially change the predicted tissue exposure, target engagement, or systemic disposition. That claim depends on the relevant time scale, length scale, and endpoint.

Transport, Uptake, and Dimensionless Scales

Consider a one-dimensional tissue region of length \(L\). Let \(c(x,t)\) be the free interstitial antibody concentration at position \(x\), \(D\) the effective tissue diffusion coefficient, \(v\) the interstitial convective velocity in the positive \(x\) direction, and \(k\) a local first-order removal rate. Here, removal is a simplified representation of unsaturated binding-associated uptake or another approximately linear sink; it is not a general description of saturable target binding.

With constant \(D\), \(v\), and \(k\), conservation of antibody gives the advection-diffusion-reaction equation. The flux is \(J=vc-D\,\partial c/\partial x\); conservation states that the change in flux with position equals the negative local removal rate.

This form assumes a continuum tissue description, constant transport coefficients over the modeled region, and a local sink proportional to free concentration. Real antibody systems may violate the last assumption when binding saturates, target abundance varies, or bound antibody dissociates appreciably. Those situations require more detailed local binding states or an empirically justified effective sink.

Two dimensionless groups organize the transport comparison. The diffusion-reaction group, \(Da_D\), compares the time scale for diffusion across the tissue, \(L^2/D\), with the removal time scale, \(1/k\). The Péclet number, \(Pe\), compares convection with diffusion. Their ratio gives a convection-reaction scale.

Small \(Da_D\) suggests diffusion can redistribute antibody across the tissue faster than the sink removes it. Small \(Da_C\) suggests convection carries antibody across the region faster than removal depletes it along that flow path. These are screening scales, not universal cutoffs. A large \(Pe\) does not by itself justify a well-mixed model: it means convection is important relative to diffusion, and a sufficiently large \(Da_C\) can still produce a pronounced downstream decrease.

A Steady Diffusion-Reaction Gradient

A simple slab calculation gives a direct link between reaction strength and spatial variation. Assume steady state, diffusion only, constant \(D\), and a linear sink \(kc\). Set the concentration at the supplied boundary to \(c(0)=c_0\), and assume zero diffusive flux at the far boundary, \(c'(L)=0\). These boundary conditions describe a slab supplied at one side with no diffusive escape through the other side; they are a deliberately simplified test of intratissue depletion.

For \(k>0\), define \(q=\sqrt{k/D}\). The general solution can be expressed relative to the zero-gradient boundary as \(c(x)=A\cosh(q(L-x))+B\sinh(q(L-x))\). Applying \(c'(L)=0\) sets \(B=0\). Applying \(c(0)=c_0\) then gives \(A=c_0/\cosh(qL)\). Thus the far-boundary concentration relative to the supplied concentration is:

When \(\phi\) is small, \(\cosh(\phi)\) is close to 1 and the profile is nearly flat. When \(\phi\) is large, the far side is substantially depleted. This result is exact only for the stated steady, one-dimensional, diffusion-only model with a linear sink and those boundary conditions; convection, time dependence, variable target density, and different tissue geometry alter the profile.

Permeability and Intratissue Transport Are Different Bottlenecks

Antibody must cross the vascular interface before it can distribute through interstitium. Low vascular permeability can limit tissue entry even if interstitial diffusion is rapid. Conversely, high permeability does not guarantee uniform tissue concentration if internal transport is slow relative to uptake. A whole-tissue compartment may represent total exchange adequately while still missing a consequential gradient inside the tissue.

A useful resistance comparison can be made for a planar, steady, diffusion-only path without reaction. Let \(P\) be the vascular permeability in cm/s. The interface conductance per area is \(P\); the conductance for diffusion across a tissue distance \(L\) is \(D/L\). Their reciprocal resistances are \(1/P\) and \(L/D\), respectively, both in s/cm. This comparison identifies the larger transport resistance; it is not a substitute for a spatial model when sinks or gradients matter.

Convection and Heterogeneous Target Uptake

When convection dominates and diffusion is negligible, a steady one-dimensional model with local first-order removal has \(v\,dc/dx=-k(x)c\). This approximation assumes positive, constant velocity, no important axial diffusion, and a sink that remains linear in free concentration. Its solution makes the role of spatial target heterogeneity explicit:

The downstream concentration depends on cumulative uptake along the path, but the concentration at intermediate positions depends on where high-uptake regions occur. A single average \(k\) may reproduce a final outlet concentration under these restrictive conditions while misrepresenting local concentrations and target exposure.

Evidence, Diagnostics, and Model Choice

Spatial resolution is most defensible when there is evidence that spatial variation is both present and relevant. Useful evidence includes imaging or tissue sampling that shows gradients, known localization of target or uptake, and systematic differences in exposure between vascular-adjacent and more distant regions. Average tissue concentration alone is weak evidence about uniformity: multiple spatial profiles can have the same average.

A practical model comparison begins with a well-mixed representation and asks whether its assumptions are credible for the intended prediction. Estimate or bound the relevant transport and uptake scales, compare the model with spatial observations where available, and examine whether predictions for the decision endpoint change when spatial structure is added. A spatial model that cannot be constrained by the available observations can produce a more detailed but less reliable answer.

Diagnostics for a potentially inadequate well-mixed model include persistent underprediction in distal tissue regions, systematic mismatch between spatially sampled concentrations, or failure to reproduce local target engagement while matching whole-tissue averages. In a spatial simulation, inspect whether concentration profiles change plausibly with permeability, diffusion, convection, and uptake. Unexpected discontinuities, negative concentrations, or strong dependence on arbitrary spatial discretization can indicate implementation or resolution problems rather than biology.

The main alternative is not always a full spatial model. A well-mixed compartment is economical and may be appropriate when internal mixing is rapid relative to removal and the endpoint is an average. A model with several lumped tissue regions can represent a coarse proximal-distal gradient at lower computational cost, but its regions and exchange rates still require justification. A continuous diffusion-convection model is more mechanistic, but demands more data and assumptions about geometry, transport, and local sinks.

Common Mistake and Limitations

• Wrong reasoning: “The tissue-average concentration matches the data, so the tissue must be well mixed.” Correction: An average does not identify the spatial profile. Check gradients or whether the endpoint depends on local rather than average exposure.
• Wrong reasoning: “High vascular permeability means the antibody is uniform throughout the tissue.” Correction: Permeability controls transfer across the vascular boundary; diffusion, convection, and local uptake determine internal distribution.
• Wrong reasoning: “A large Péclet number means the tissue is well mixed.” Correction: A large \(Pe\) indicates convection relative to diffusion. If \(Da_C=kL/v\) is also large, removal can produce a substantial gradient along the flow path.
• Wrong reasoning: “The first-order sink is always a valid representation of target binding.” Correction: It is a simplifying approximation. Saturation, reversible binding, and spatially variable target abundance may require explicit local binding or uptake states.

Dimensionless groups are screening tools, not universal pass-fail thresholds. Effective diffusion and convection can vary among tissues and with physiological state, while target heterogeneity can occur at scales smaller than an organ-level model. Spatial models also have limitations: boundary conditions may be uncertain, spatial measurements may be sparse, and adding compartments or grid points does not guarantee identifiability. Resolution should be matched to the biological scale and the data, rather than increased automatically.