Why Thermal Shifts Are Evidence, Not Occupancy Measurements
A compound can alter a protein’s resistance to heat-induced loss from the soluble fraction. CETSA-type assays use this change as a readout: cells, tissue, or lysate are exposed to compound, heated under controlled conditions, and the remaining soluble target is quantified. A reproducible concentration-dependent change can support the conclusion that the compound engages the target in that assay system. The central interpretive limit is that the assay measures a thermal-solubility response, not the fraction of target molecules occupied by compound.
The distinction matters because binding can alter protein stability, but it need not do so detectably; conversely, a change in soluble target can arise indirectly through protein complexes, post-translational state, localization, or abundance. Even when the response is caused by binding, its size need not be proportional to occupancy. Thermal-shift assays are therefore valuable evidence of engagement, but a shift alone does not establish an affinity constant or a numerical occupancy estimate.
What the Assay Measures
A typical temperature-response experiment exposes matched samples to a range of temperatures after treatment with vehicle or compound. The soluble target is quantified by an appropriate method, such as immunodetection or targeted proteomics. Because the raw signal can also depend on starting target abundance, it is useful to compare each heated sample with a matched unheated sample from the same treatment condition.
Let \(S(T,C)\) denote the target signal after heating at temperature \(T\) and treatment concentration \(C\). Let \(S(T_0,C)\) be the corresponding signal in an unheated or minimally heated sample. A normalized soluble fraction can be written as
This normalization helps separate heat-dependent loss from differences in the amount of target present before heating. It does not remove every source of bias: for example, treatment-dependent changes in antibody recognition or sample recovery may still affect the signal.
An operational midpoint, often denoted \(T_m\) in reporting, is the temperature at which the fitted normalized response is halfway between its upper and lower plateaus. It is an assay-specific summary of the transition, not necessarily a thermodynamic melting temperature. The difference between compound and vehicle midpoints is
A positive \(\Delta T_m\) indicates a shift toward higher temperatures under the stated conditions. A negative value indicates a shift toward lower temperatures. The sign alone does not establish whether the compound binds directly to the target.
Why a Thermal Response Does Not Determine Occupancy
A simple mixture argument shows the limitation. Suppose that at a fixed temperature the unbound target has probability \(q_U\) of remaining soluble, and the bound target has probability \(q_B\). Let \(\theta\) be the fraction of target bound. If the assay signal is proportional to soluble target and bound and unbound molecules are detected equally, the expected soluble fraction is
The expression follows by adding the soluble contributions from the unbound fraction, \(1-\theta\), and the bound fraction, \(\theta\). Even in this simplified case, \(F\) cannot determine \(\theta\) unless the relevant survival probabilities are known and stable. In real assays, those probabilities can depend on temperature, sample matrix, protein state, and processing. If \(q_B=q_U\), binding can occur with no thermal signal at all. If an indirect process changes \(q_U\), a signal can change without a change in direct target binding.
For the same reason, the concentration giving half of a maximal thermal response is an apparent assay EC50, not automatically a binding \(K_D\), an IC50 from a displacement experiment, or the concentration for 50% occupancy. It depends on the assay temperature, the response mechanism, incubation time, free compound concentration, and the shape of the readout.
Worked Examples
Worked Example: Calculating an Operational Thermal Shift
In a hypothetical temperature-response experiment, normalized soluble fractions are measured for vehicle and compound-treated samples. The vehicle curve has \(F=0.75\) at \(49^\circ\text{C}\), \(F=0.50\) at \(51^\circ\text{C}\), and \(F=0.25\) at \(53^\circ\text{C}\). The compound curve has the same three fractions at \(51^\circ\text{C}\), \(53^\circ\text{C}\), and \(55^\circ\text{C}\), respectively. Estimate the operational midpoints and the shift.
Vehicle midpoint. The normalized response is 0.50 at \(51^\circ\text{C}\), so \(T_{m,\text{vehicle}}=51^\circ\text{C}\).
Compound midpoint. The normalized response is 0.50 at \(53^\circ\text{C}\), so \(T_{m,\text{compound}}=53^\circ\text{C}\).
Shift.
The compound condition shows a \(2^\circ\text{C}\) shift toward higher temperatures in this hypothetical assay. This supports a change in target thermal stability under the tested conditions. It does not show that target occupancy is 50% at any particular concentration: the midpoint here is a temperature midpoint, not a concentration midpoint or occupancy estimate.
Worked Example: Interpreting a Concentration-Dependent Signal
At a fixed heating temperature, a hypothetical assay has normalized signal \(S_0=0.20\) with vehicle and a plateau signal \(S_{\max}=0.80\) at high compound concentration. Suppose the response is adequately summarized over the tested range by
where \(C\) is the nominal compound concentration in micromolar and \(EC_{50,\text{app}}=5\,\mu\text{M}\). Calculate the signal at \(C=1\,\mu\text{M}\), \(5\,\mu\text{M}\), and \(20\,\mu\text{M}\).
The response range is \(S_{\max}-S_0=0.80-0.20=0.60\). At \(1\,\mu\text{M}\),
At \(5\,\mu\text{M}\),
At \(20\,\mu\text{M}\),
The signal rises with concentration and reaches the midpoint of its response range, \(0.50\), at \(5\,\mu\text{M}\). The fitted value is an apparent assay EC50 in this fixed-temperature setup. It is not a measured \(K_D\) or a direct estimate of 50% target occupancy. Interpretation also assumes that the chosen response form describes the data adequately and that the concentrations used are available to the target rather than substantially depleted, precipitated, or otherwise unavailable.
Worked Example: Detecting an Abundance-Related Artifact
In a hypothetical experiment, vehicle-treated samples give an unheated target signal of 80 units and a heated signal of 48 units. Compound-treated samples give an unheated signal of 120 units and a heated signal of 72 units. A report based only on heated signals might claim that the compound increased the soluble target signal from 48 to 72 units. Check the normalized fractions.
For vehicle,
For compound,
Both conditions retain 60% of their unheated signal after heating. The raw heated signal is 50% higher under compound treatment, since \((72-48)/48=0.50\), but the normalized thermal response is unchanged. The data are consistent with more target signal before heating, not with greater heat resistance. Unheated controls are therefore essential when interpreting changes in the amount of soluble target.
Controls, Diagnostics, and Failure Modes
A credible thermal-shift interpretation depends on controls that distinguish a target-specific, heat-dependent response from changes in sample handling or detection. The exact design depends on whether intact cells, tissue, or lysate are used, but several checks are broadly useful.
- Vehicle-matched samples: Keep solvent concentration and handling comparable across conditions. A vehicle effect can alter the thermal curve and confound the compound contrast.
- Unheated samples: Measure target signal before the heat challenge for each treatment. A treatment-related change in starting abundance should not be mistaken for increased heat stability.
- Temperature coverage: Use enough temperatures to characterize both plateaus and the transition. If all measurements lie on one plateau, a midpoint and shift are poorly determined.
- Concentration coverage: Include concentrations spanning the baseline through the response plateau where feasible. A rising response without a plateau may not identify a stable apparent EC50.
- Replicate and detection checks: Confirm that the signal is reproducible and within the quantitative range of the detection method. Saturated or near-background signals can distort apparent curve shape.
- Specificity controls: Where feasible, compare with a suitable inactive analog, a target-negative or target-reduced system, or a validated competing ligand. Each control addresses a different alternative explanation; none should be assumed to be informative without checking its own effects.
- Orthogonal evidence: A method based on a different measurement principle can help determine whether the result reflects target engagement rather than an assay-specific change in solubility or detection.
Potential warning signs include inconsistent replicate curves, a concentration response that tracks compound precipitation or cytotoxicity, large shifts in unheated signal, and a fitted midpoint outside the measured temperature range. A change in curve shape also deserves attention: a single midpoint can conceal multiple transitions or a broadened response. These patterns do not automatically invalidate the experiment, but they weaken a simple “stabilization” interpretation and call for investigation.
How to Interpret the Evidence and Its Alternatives
Thermal-shift data are best reported as assay observations first: the sample system, incubation conditions, temperature or concentration range, normalized signal definition, and estimated shift or apparent concentration-response summary. Then state the inference at the appropriate level: the compound altered the target’s thermal-solubility response under those conditions, consistent with engagement. Avoid converting that observation directly into occupancy or affinity without independent evidence and a justified measurement model.
| Approach | Primary evidence | Key interpretive limit |
|---|---|---|
| CETSA-type thermal shift | Change in target thermal solubility after compound treatment | Indirect engagement evidence; response is not occupancy by definition |
| Radioligand displacement or tracer-based occupancy assay | Competition or change in tracer binding under its assay conditions | Requires appropriate tracer, binding, and sample assumptions |
| Functional assay | Change in a downstream or proximal biological response | Response can depend on steps beyond target binding |
These approaches answer related but different questions. A thermal shift reports a change in protein behavior under heat challenge. A displacement or tracer assay can provide more direct binding or occupancy evidence when its assumptions are met. A functional assay reports a biological consequence, which may not track occupancy one-to-one. Agreement across approaches strengthens a mechanistic interpretation; disagreement can reveal that the assays differ in system, timing, sensitivity, or what they measure.
Thermal shifts also have limitations even when carefully controlled. Some genuine binders produce little detectable shift. Some target states may not yield a clear transition. In intact cells, compound access and the intracellular environment affect the result; in lysates, cellular context and ongoing processes differ. The method is consequently most useful as one component of an evidence chain, not as a universal substitute for direct occupancy measurement.
Common Mistake
Check Your Understanding
Answer each question, then compare your reasoning with the solutions.
- A compound changes the operational midpoint from \(48^\circ\text{C}\) to \(50.5^\circ\text{C}\). What is the shift, and what does it establish?
- At a fixed temperature, the vehicle signal is 0.15, the plateau is 0.75, and the apparent EC50 is \(3\,\mu\text{M}\). Using the one-site response form shown above, what signal is predicted at \(3\,\mu\text{M}\)?
- Why is an unheated sample needed for each treatment condition?
- Can an unchanged thermal curve rule out target binding? Explain.
Solutions. (1) \(\Delta T_m=50.5-48=2.5^\circ\text{C}\). This establishes a higher operational thermal midpoint under the tested conditions, not a particular occupancy or proof of direct binding. (2) At \(C=EC_{50,\text{app}}=3\,\mu\text{M}\), the concentration term is \(3/(3+3)=0.5\). Thus \(S=0.15+(0.75-0.15)(0.5)=0.15+0.30=0.45\). (3) It provides the treatment-specific starting signal, allowing heat-dependent solubility to be distinguished from changes in target abundance or detection signal before heating. (4) No. Binding may not change thermal solubility detectably, so an unchanged curve does not by itself exclude engagement.