Selecting T-Cell Biomarkers to Separate Mechanisms

Why Biomarker Selection Is a Mechanistic Problem

A biomarker panel is useful when its measurements distinguish plausible explanations for an observed response or resistance. A rising blood T-cell count, for example, establishes that cells expanded in circulation; by itself, it does not show whether tumor-reactive cells were primed, entered the tumor, retained cytotoxic capacity, or reached tumor cells in the tissue. The same tumor outcome can therefore be consistent with different failures along the response pathway.

For checkpoint inhibitors and CAR T-cell therapies, organize candidate measurements around four linked processes: priming of tumor-reactive T cells, infiltration into tumor tissue, exhaustion or other loss of functional state, and cytotoxic function. The aim is not to measure the largest possible number of markers. It is to choose measurements that address competing mechanisms, in the compartments where those mechanisms occur, and at times when their differences can be observed.

Map Each Measurement to a Biological Process

For a priming question, measurements should reflect activation of antigen-specific T cells and the environment that supports their activation. Examples include activated antigen-presenting cells in a draining lymph node, proliferation of tumor-antigen-specific T cells, and early expansion of tumor-reactive clonotypes. A rise in total circulating lymphocytes is less specific: it may reflect cells unrelated to the tumor antigen or changes in other cell populations.

For infiltration, prioritize tumor-tissue measurements. These can include CD8 T-cell density per unit area, the abundance of tumor-reactive clonotypes in a biopsy, and their spatial distribution relative to tumor cells. Blood measurements cannot substitute for tumor sampling: a clone can expand in blood yet fail to enter tumor tissue. Even a high tissue density may be incomplete evidence of access if the cells are concentrated in stromal regions rather than near tumor cells.

For exhaustion or functional state, use a combination of state-associated features and functional measurements. Inhibitory receptors such as PD-1 can be expressed during activation as well as in dysfunctional states; expression of one receptor is not, by itself, a diagnosis of exhaustion. Co-expression patterns, transcriptional features, duration of stimulation, and performance in a controlled functional assay provide complementary evidence. For cytotoxic function, possible measurements include target-cell killing, degranulation, and effector molecules such as granzyme B. Whenever possible, distinguish the number of cells present from the function of each cell.

These measurements should be assigned to a compartment and a time point before collecting data. Priming is often most directly assessed in a draining lymph node or through a validated circulating surrogate; infiltration and spatial access require tumor tissue; systemic expansion can be followed in blood. Baseline and early on-treatment samples can reveal initiating changes, while later samples can assess persistence and tumor response. A single late biopsy may miss an early priming event.

Use Competing Predictions to Choose the Panel

Begin by naming the explanations that could account for the outcome. For example, a priming-limited response predicts weak generation or expansion of tumor-reactive T cells, while an effector-limited response can include substantial priming but inadequate function in the tumor. A trafficking limitation predicts that tumor-reactive cells may be detectable in blood but scarce, or poorly positioned, in tumor tissue. These predictions suggest different measurements; simply adding more markers from the same compartment may not resolve the alternatives.

A useful formal check is whether candidate measurements respond differently to changes in the mechanisms of interest. Let \(\theta\) represent a vector of mechanistic quantities and \(y\) a vector of measured biomarkers. Near a reference condition, small changes can be approximated by:

Here, \(J\) is the local sensitivity matrix, whose entry \(J_{ij}\) describes how much biomarker \(i\) is expected to change for a small change in mechanism \(j\). If two mechanisms produce nearly identical changes in every selected measurement, those measurements do not distinguish them well. In the exact linear case, if two columns of \(J\) are identical, the corresponding mechanisms cannot be separated using those measurements alone. A measurement that responds differently to the two mechanisms can improve discrimination.

Precision matters as well as biological sensitivity. If two explanations predict values that differ by less than the usual measurement variability, that biomarker may be a weak discriminator in practice. A simple planning score for two predictions, assuming independent measurement errors with known standard deviations, is the sum of squared differences divided by the corresponding variances. This score helps compare proposed panels; it is not a hypothesis test or proof that one mechanism is true.

Separate Abundance, Location, State, and Function

An important design principle is to avoid treating a single quantity as a proxy for several biological steps. The number of T cells in blood is not the number in the tumor. The density of cells in a biopsy is not necessarily the fraction that recognizes tumor antigen. The fraction expressing an exhaustion-associated marker is not a direct measurement of killing. Whenever the mechanistic question involves a transition between steps, measure on both sides of that transition if feasible.

Account for Time, Specificity, and Normalization

Sampling time changes what a biomarker can answer. A transient early increase in antigen-specific proliferation may be missed by a late sample; conversely, a late functional deficit cannot by itself show whether inadequate priming occurred earlier. A practical sequence might include baseline, an early sample after treatment begins, and a later sample aligned with the tumor assessment. The precise schedule depends on the expected biology and sample feasibility.

Specificity also matters. If the claim concerns tumor-reactive cells, measurements should distinguish those cells from total T cells wherever possible, for example through antigen-specific assays or tracking relevant clonotypes. When this is not feasible, state the limitation clearly: a nonspecific marker can describe a broad immune change but cannot alone attribute that change to tumor recognition.

For functional assays, report conditions that affect interpretation, including the target cells, effector-to-target ratio, stimulation, and assay duration. To distinguish a change in cell abundance from a change in per-cell function, normalize functional output to viable effector-cell number or use a fixed ratio. Bulk cytokine concentration without cell-count information can increase because there are more cells, because each cell produces more, or both.

The measurement scale should match the question. Absolute cell density per tissue area, percentage of a cell population, and abundance per milliliter are not interchangeable. Use consistent units and, for longitudinal comparisons, the same assay and processing where possible. A result close to the assay’s detection limit, a large difference between replicate measurements, or disagreement between tissue and blood should prompt review of assay performance and sampling before a mechanistic conclusion is drawn.

Comparing Biomarker Strategies and Their Limits

A single-marker strategy is efficient and can be appropriate when the marker directly measures a well-defined event, but it is vulnerable to ambiguity: PD-1 alone does not separate activation from dysfunction, and blood expansion alone does not establish tumor infiltration. A compartment-specific panel combines complementary measurements, such as node priming, blood expansion, and tissue localization. It is more informative about where a response may be limited, but requires more samples and careful alignment of sampling times. A functional assay asks whether cells perform a defined task, but its result depends on assay conditions and may not reproduce in-tumor performance.

No panel removes all uncertainty. Tumors are spatially heterogeneous, biopsies sample only a small region, and treatment can alter the meaning of a marker over time. Measurements can also be correlated, so the independence assumption in the simple separation score above may be inappropriate. In that case, a covariance-aware comparison is preferable, but estimates of the covariance must themselves be reliable. Biomarker patterns should be interpreted alongside tumor outcomes and the known intervention; they remain evidence about mechanisms rather than direct proof of causation.

Common Pitfall

Wrong reasoning: “CAR T cells expanded five-fold in blood, so they must have entered the tumor and remained functional; the weak response must be caused by tumor resistance.” The conclusion assigns tissue entry, cell state, and causation to one blood measurement that directly establishes none of them.

Correction: Treat blood expansion as evidence of systemic abundance. Add tumor measurements for infiltration and spatial access, and a controlled functional measurement for cytotoxic capacity. Even if these measurements suggest a likely bottleneck, describe the result as consistent with that mechanism unless the study design supports a causal claim.