Why the Biomarker Category Matters
A Phase I biomarker result can be encouraging without answering the question people most want to ask: will the drug help patients? A measurement may show that a drug reached and interacted with its intended target, that a biological process changed, or that a biomarker can stand in for a clinical endpoint. These are different claims, supported by different kinds of evidence.
The distinctions matter when interpreting early data. A marker of target engagement can support the claim that the drug interacted with its intended target under the conditions measured. A pharmacodynamic (PD) marker can support the claim that a biological response occurred. Neither claim alone establishes that patients feel better, function better, or live longer. A surrogate endpoint makes a stronger claim: it is used in place of a clinical endpoint, and its suitability depends on evidence that goes beyond observing a biomarker change.
These labels describe different aspects of a measurement, and they are not always mutually exclusive. For instance, a biomarker that directly measures target activity might provide evidence of target engagement and also be described as pharmacodynamic because it registers a drug-induced biological change. “Surrogate,” in contrast, describes a proposed role in evaluating clinical benefit. Calling a measurement a biomarker does not, by itself, establish any of these roles.
What Each Category Can Support in Phase I
A target-engagement result is most directly informative about whether the drug interacted with the intended target in the sampled system and at the measured time. Depending on the measurement, it may indicate binding, occupancy, or another direct measure of target interaction. It does not necessarily establish that the target’s function changed in the intended direction, that a downstream pathway responded, or that a patient benefited.
A PD biomarker measures a biological response associated with drug administration. It may be close to the target, such as a change in a target-controlled biochemical product, or farther downstream, such as a change in a pathway marker. A response may be consistent with the proposed mechanism, but it can also arise through other biological routes. Conversely, no observed PD change does not automatically prove there was no target interaction: the marker may not reflect the relevant process, or the measurement may not capture the response.
In Phase I, these biomarkers can help characterize whether a proposed mechanism is biologically active in humans and can inform further development decisions. Their interpretation must remain tied to what was measured. A positive target-engagement result supports evidence of target interaction; a positive PD result supports evidence of a measured biological response. Neither alone establishes an improvement in clinical outcomes.
A surrogate endpoint is different because it is intended to substitute for a clinical endpoint. That substitution requires evidence that changes in the biomarker reliably predict the treatment effect on the clinical outcome in the relevant context. Biological plausibility is useful, but plausibility alone is not sufficient. Nor does a correlation between a biomarker and outcome among individuals establish that a treatment-induced change in the biomarker captures the treatment’s effect on the outcome.
| Category | What is measured or claimed? | What a Phase I result can support | What it does not establish by itself |
|---|---|---|---|
| Target engagement | Direct evidence of drug interaction with the intended target | Evidence of target interaction in the measured setting | Downstream response or clinical benefit |
| Pharmacodynamic biomarker | A biological response associated with the intervention | Evidence that the measured biological process changed | That the change improves a clinical outcome |
| Surrogate endpoint | A biomarker proposed for use instead of a clinical endpoint | Only what the supporting evidence justifies in the relevant context | Validity based solely on mechanism or association |
Why a PD Response Does Not Prove Clinical Benefit
A biological response is not the same thing as a beneficial clinical response. The chain of reasoning often proposed is: the drug engages its target; target engagement changes a pathway; the pathway change improves a clinical outcome. Evidence for an early link in that chain does not automatically establish the later links. The pathway may not be sufficiently important to the outcome, another process may offset the change, or the response may have unwanted effects.
A simple relationship shows why the inference is not automatic. Let \(B\) denote a biomarker and \(Y\) a clinical outcome. Suppose, only for illustration, that the outcome change is related to the biomarker change by a coefficient \(\beta\) plus other influences:
Here, \(\Delta U\) collects changes in influences on the clinical outcome not represented by the biomarker. Even when \(\Delta B\) is clearly nonzero, \(\Delta Y\) need not indicate benefit: the biomarker may not be meaningfully related to the outcome (\(\beta\) may be near zero), or other influences may offset the effect. This simple equation is not a test for surrogate validity. It illustrates why observing a biomarker response alone cannot establish the clinical consequence.
The distinction also clarifies a common reasoning error: “The marker is in the proposed disease pathway, so changing it must improve the disease.” The marker’s place in a plausible pathway can motivate investigation. It does not show that changing the marker causes a clinical improvement or that it captures the effects of treatment on the outcome. Surrogate status therefore needs evidence relevant to the intended treatment, population, clinical outcome, and context of use.
Worked Example: Calculating a Target-Engagement Estimate
In a hypothetical Phase I study, a calibrated measurement estimates that 72 of 120 measurable target molecules in a sampled system are occupied by the investigational drug at the specified assessment. For this example, assume that the occupied and total counts are measured on a comparable basis and that the estimate is suitable for calculating the occupied fraction.
The estimated occupied fraction is the number occupied divided by the total measurable target:
Expressed as a percentage, the estimate is \(0.60 \times 100 = 60\%\). In this hypothetical measurement, the result supports evidence that approximately 60% of the measured target was occupied in the sampled system at the assessment. It does not show that the target’s downstream function changed, that the relevant tissue had the same occupancy, or that patients experienced clinical benefit. Those are separate questions.
Worked Classification and Interpretation
Worked Example: Classifying a Biomarker Panel
A hypothetical first-in-human study measures three indicators after dosing: (1) the fraction of a receptor occupied by drug, (2) the concentration of a downstream inflammatory protein, and (3) a symptom score intended to replace a clinical outcome in later trials. Classify each measurement and state what it can support.
(1) Receptor occupancy. This is a target-engagement biomarker because it directly measures drug interaction with the intended receptor. A change in the measured occupied fraction can support target interaction in the assessed system. It does not, by itself, establish downstream functional change or clinical benefit.
(2) Downstream inflammatory protein. This is a PD biomarker if it is used to indicate a biological response associated with drug administration. A measured change supports the claim that this protein changed under the study conditions. It does not prove that the change was caused only by the proposed target mechanism or that it improves a clinical outcome.
(3) Symptom score proposed as a substitute. Its intended role is that of a surrogate endpoint, but intention is not validation. In this Phase I study, measuring the score does not establish that it can stand in for a clinical outcome. Evidence supporting that use must show that treatment effects on the score reliably reflect treatment effects on the relevant clinical endpoint in the intended context.
The classifications are based on what each measurement represents and how it is intended to be used—not simply on whether it changes after dosing. If the receptor measurement also reflects a functional biological response, it may have both target-engagement and PD relevance. That overlap does not make it a surrogate.
Worked Example: Comparing Biomarker and Clinical Outcomes
In a hypothetical randomized Phase I study, 10 participants receive the investigational drug and 10 receive placebo. A pre-specified biomarker response is observed in 8 drug recipients and 2 placebo recipients. Clinical improvement is observed in 4 drug recipients and 3 placebo recipients. Calculate the observed response proportions and explain what can be concluded.
For the biomarker, the drug-group proportion is \(8/10 = 0.80\), or 80%, and the placebo-group proportion is \(2/10 = 0.20\), or 20%. The observed difference is:
Thus, the observed biomarker-response proportion is 60 percentage points higher in the drug group. For clinical improvement, the corresponding proportions are \(4/10 = 0.40\) and \(3/10 = 0.30\), giving an observed difference of:
The hypothetical data show a larger observed separation in biomarker response than in clinical improvement. They support describing those observed proportions; with only 10 participants per group, they do not establish a precise treatment effect. More importantly for this tutorial, the biomarker separation does not establish clinical benefit or validate the biomarker as a surrogate. A small Phase I study may be informative about biological activity while remaining inadequate to establish a biomarker’s ability to substitute for a clinical outcome.
Evidence, Diagnostics, and Limitations
Interpretation should start by asking what the measurement directly records. Is it evidence of drug–target interaction, a biological response, or a clinical outcome? Then ask what inference the result is being used to support. A claim that reaches beyond the measurement—for example, inferring patient benefit from occupancy alone—needs additional evidence.
Several warning signs should prompt restraint:
- A category is assigned from the result rather than the measurement. A marker does not become target-engagement evidence just because it changes after dosing; the measurement must indicate direct target interaction.
- A pathway explanation is treated as proof. A plausible mechanistic link can support a hypothesis, but it does not establish that the measured response improves a clinical outcome.
- Association is treated as surrogate validation. A biomarker may be associated with disease severity or outcome without reliably capturing treatment effects on that outcome.
- The measured context is overgeneralized. Evidence in one biological system, population, or setting does not automatically justify the same conclusion in another.
- Overlapping roles are forced into one label. A marker may provide both target-engagement and PD information, while still lacking evidence for surrogate use.
The main alternative to using a biomarker as a substitute is to retain a clinical endpoint as the measure of clinical benefit, while using the biomarker as supportive evidence about biological activity. This distinction is especially useful in early development: a biomarker can be informative even when it is not a validated surrogate. The limitation is that a supportive marker may not resolve whether the intervention produces a meaningful clinical effect. That question requires evidence on the clinical outcome itself or sufficiently strong evidence to justify the proposed surrogate role.
Check Your Understanding
For each question, identify the appropriate claim and its limits. Brief solutions follow the questions.
- A measured drug-bound fraction of a molecular target increases after dosing. What category of evidence does this most directly provide, and what does it not establish?
- A downstream protein concentration changes after treatment. What claim can this support, and why is it not automatically evidence of clinical benefit?
- A biomarker is associated with disease severity across patients. Does that association alone establish that the biomarker is a valid surrogate endpoint? Explain.
- In a hypothetical study, a response is observed in 15 of 25 drug recipients and 10 of 25 placebo recipients. Calculate the observed difference in proportions and state one conclusion it does not justify.
Solutions. (1) It most directly provides target-engagement evidence, assuming the measurement indicates drug interaction with the intended target; it does not establish downstream response or benefit. (2) It can support a PD response, but that response may not improve an outcome that matters to patients. (3) No. Association across patients does not show that treatment effects on the biomarker reliably represent treatment effects on the clinical endpoint. (4) The proportions are \(15/25 = 0.60\) and \(10/25 = 0.40\); the observed difference is \(0.60 - 0.40 = 0.20\), or 20 percentage points. This calculation does not establish clinical benefit or surrogate validity.