🌡️ Thermal Shift Assay / Differential Scanning Fluorimetry (TSA/DSF)

Figure 1 schematic explaining thermal shift assay and differential scanning fluorimetry: folded protein with weak dye signal, heating-driven unfolding exposing hydrophobic regions, fluorescent dye binding, melt curve, and ligand-induced Tm shift

The thermal shift assay is one of the simplest, cheapest, and fastest ways to check if something stabilizes your protein. Mix protein with a hydrophobic fluorescent dye (SYPRO Orange), heat it up, and watch the fluorescence. As the protein unfolds, its hydrophobic core is exposed, the dye binds → fluorescence increases → sigmoidal melt curve → you get T_m. Add a ligand that stabilizes the protein → T_m shifts up → ΔT_m is your readout.

Dye-based TSA runs on a compatible real-time qPCR machine with suitable excitation/emission filters. It's widely used for buffer/excipient screening in formulation development, fragment screening in drug discovery, and general "is my protein folded?" QC checks. An entire 96-well plate typically runs in 30–60 minutes with low microgram protein per well.

What a single TSA shift doesn't do: ΔT_m is NOT a K_D. A +3°C shift from compound A and a +3°C shift from compound B does not mean they have the same affinity — they might bind different sites, have different enthalpic contributions, or stabilize different conformational states. Single-dose TSA is a rank-ordering and triage tool, not a direct quantitative affinity measurement. Dose-response thermal-shift analysis can estimate K_D, but only with model assumptions and good controls. For routine K_D, use SPR, MST, ITC, or FP.

Key Physics Concepts

🔥

Protein Unfolding Thermodynamics

Proteins unfold when thermal energy overcomes stabilizing forces. The two-state model assumes:

N ⇌ U

K(T) = exp(−ΔG(T) / RT)

At T_m: K = 1, ΔG = 0, f_u = 0.5. Simplified van't Hoff (ΔC_p = 0):

K(T) = exp[−(ΔH_u/R)(1/T − 1/T_m)]

ΔH_u > 0 (unfolding endothermic), so K > 1 above T_m and K < 1 below. Valid within ±15°C of T_m. ΔH_u ≈ 100–600 kJ/mol for typical 15–50 kDa globular proteins.

💡

SYPRO Orange — The Reporter Dye

SYPRO Orange is an environmentally sensitive fluorophore:

In aqueous solution: fluorescence is quenched (low quantum yield)
In hydrophobic environments: fluorescence is bright (high quantum yield)
λ_ex ≈ 470 nm · λ_em ≈ 570 nm (ROX/TAMRA filter)
5000× stock in DMSO; use 5–10× final concentration

Folded protein → hydrophobic core buried → low fluorescence. Unfolded protein → core exposed → dye binds → high fluorescence. At higher temperatures, aggregation, precipitation, dye redistribution, and thermal quenching can make the signal plateau or decrease.

🔬

Nano-DSF — Label-Free Intrinsic Fluorescence

Nano-DSF eliminates the dye by monitoring intrinsic tryptophan fluorescence:

Buried Trp (folded): λ_em ≈ 330 nm (blue-shifted, non-polar)
Exposed Trp (unfolded): λ_em ≈ 350 nm (red-shifted, aqueous)
Ratio R = F₃₅₀ / F₃₃₀ often changes upon unfolding — it may increase, decrease, or be less informative than a single wavelength
Ratiometric readout is less sensitive to concentration and intensity drift than raw fluorescence, but not immune to signal-quality artifacts

Instruments: NanoTemper Prometheus NT.48, Prometheus Panta, Unchained Labs UNit/UNCLE. Requires only 10 µL at 0.1–1 mg/mL.

Interactive TSA/DSF Simulator

Three modes: melt curve physics, ΔTm ligand screening, and nano-DSF (350/330 ratio).

Tm(melting temperature)60 °C

ΔH_m(unfolding enthalpy)300 kJ/mol

F_folded(baseline fluorescence)200 RFU

F_unfolded(max fluorescence)5000 RFU

Readout

T_m60 °C

Transition width (10–90%)13 °C

ΔH_m300 kJ/mol

CooperativityNormal

Derivative convention: dF/dT is plotted directly — positive peak at T_m. Some qPCR software plots −dF/dT (designed for DNA melts where fluorescence decreases). For protein DSF, −dF/dT gives a trough at T_m. If you see a trough in your software, invert the sign or read the minimum as T_m.

Fluorescence vs Temperature

How DSF Works — The Measurement

🔬 The qPCR Protocol (Dye-Based)

Prepare plate: 20 µL reactions in 96- or 384-well PCR plate. Common starting point: 1–5 µM protein, 5× SYPRO Orange, buffer of choice, ± ligand/compound.
Seal plate with optically clear adhesive film.
Run thermal ramp: 25 → 95°C at 0.5–1°C/min in qPCR machine.
Read fluorescence every 0.5–1°C (use ROX or TAMRA filter set).
Analyze: Plot fluorescence vs. T → determine T_m from derivative peak.

Typical runtime: 30–60 min per plate. Requires a real-time qPCR machine with a filter set compatible with SYPRO Orange emission.

🔬 The Prometheus Protocol (Nano-DSF)

Load capillaries: 10 µL per capillary, up to 48 samples.
No dye needed — intrinsic Trp/Tyr fluorescence.
Thermal ramp: 15 → 95°C at 1°C/min (standard).
Dual-UV detection: F₃₃₀ and F₃₅₀ simultaneously.
Back-reflection: Turbidity / reflected-light loss for aggregation or precipitation onset (T_agg).
Analyze: Plot 350/330 ratio vs. T → T_m from inflection point.

Assay Design

DSF Assay Design Checklist

Protein concentration optimization: Run a protein titration (0.5–20 µM) with fixed SYPRO Orange → find the lowest concentration that gives S/N > 3. Too high → inner filter effects; too low → poor signal.
Dye concentration optimization: Test 2.5×, 5×, 10×, 20× SYPRO Orange → find the concentration that gives good fluorescence without destabilizing the protein (dye binding can lower T_m by 1–3°C at high concentrations).
Buffer compatibility check: Run buffer-only controls with SYPRO Orange → ensure no dye fluorescence in the absence of protein. Arginine, PEG, and some other excipients interact directly with the dye.
DMSO tolerance: If screening compounds in DMSO, determine max DMSO concentration that doesn't shift T_m. Many proteins tolerate 1–2% DMSO, but the direction and size of the shift are protein-dependent. Match DMSO across all wells.
Positive control: Include a known binder with expected ΔT_m for plate-to-plate normalization.
Negative control: Protein + DMSO (no compound) → defines baseline T_m.
No-protein control: SYPRO Orange + buffer only → confirms no background signal.
Plate layout: Avoid outer wells (edge effects from thermal gradients). Randomize compound positions.

What Makes a Good TSA Target?

Protein Feature	Suitability	Why
Globular, well-folded, single domain	✅ Ideal	Clean two-state unfolding, sharp transition
Multi-domain protein	⚠️ Multiple transitions	May see 2+ T_m values — each domain unfolds independently
Intrinsically disordered	❌ Poor	Often no cooperative hydrophobic-core exposure → weak, broad, or non-classical signal
Membrane protein (in detergent)	⚠️ Tricky	SYPRO Orange binds detergent micelles → high background. Use nano-DSF or CPM
Protein with exposed hydrophobic patches	⚠️ Caution	High baseline fluorescence even when folded → compressed dynamic range
Very thermostable (T_m > 90°C)	⚠️ Limited	Transition may be beyond instrument range (95–100°C max)
Oligomeric protein	⚠️ Complex	Dissociation + unfolding → multi-step transitions, concentration-dependent T_m

Interpreting ΔT_m

The Schellman equation (quantitative relationship between ΔT_m, K_D, and [L]):

ΔT_m ≈ (R × T_m² / ΔH_m) × ln(1 + [L] / K_D)

T_m in Kelvin · ΔH_m in J/mol (unfolding enthalpy, >0) · K_D at T_m (not at 25°C)

Assumes: binding only to the native state; L_total ≈ L_free (no tight-binding depletion); ΔC_p = 0 (ignoring ΔCp introduces error of several °C for typical ΔCp ~5–10 kJ/mol/K).

Critical caveat: K_D at T_m ≠ K_D at 25°C. Protein–ligand affinity is temperature-dependent. Extracting K_D at 25°C from ΔT_m requires assumptions about the temperature dependence of binding — which you usually don't know. This is why ΔT_m should not be reported as K_D.

Common Assay Formats

🎯

Hit Identification

Screen compound library at single concentration (typically 100–500 µM). Hits = compounds with ΔT_m > cutoff (typically > 1–2°C above 3σ of DMSO control). Throughput: 96 compounds per plate, 10+ plates per day. False positive rate: ~5–15% (aggregators, fluorescent compounds, DMSO effects). Always validate hits with dose-response and orthogonal methods.

🧪

Buffer / Excipient Screening

Optimize protein stability for storage, purification, or formulation. Test pH range (4–9), salt concentrations (0–500 mM), additives (glycerol, sucrose, arginine). Higher T_m often indicates a more thermally stable formulation, but DSF is one part of a broader stability package rather than a stand-alone regulatory requirement.

🧩

Fragment Screening

Fragments (MW 150–300 Da) have weak affinity (K_D ~ 0.1–10 mM). Need high [fragment] (1–10 mM) to see ΔT_m. Typical ΔT_m: 0.5–3°C. Advantages: no labeling, no immobilization, very high throughput. Cocktail screening: Pool 5–10 fragments per well → deconvolute hits later. DMSO tolerance critical — keep ≤ 2%.

✅

Protein QC / Batch Comparison

Compare T_m between protein batches → detect stability changes that may come from aggregation, degradation, misfolding, or formulation differences. T_mshould be reproducible within the assay's validated precision. Run alongside SEC-HPLC and orthogonal biophysical assays for comprehensive QC.

Data Analysis

T_m Determination Methods

Method 1: Boltzmann Sigmoid Fit

F(T) = F_min + (F_max − F_min) / (1 + exp((T_m − T) / slope))

Four parameters: F_min, F_max, T_m, slope. Simple, works for clean two-state transitions. The slope parameter is empirical — does NOT directly yield ΔH_m. Note: slope = R·T_m²/ΔH_m (the factor-of-4 belongs to the slope of f_u, not to this Boltzmann parameter).

Method 2: First Derivative Peak

Compute dF/dT numerically (Savitzky-Golay smoothing recommended, window 5–11 points). T_m = temperature at the maximum of dF/dT (for protein DSF where F increases). If qPCR software outputs −dF/dT, T_m is the minimum. More robust for noisy data and asymmetric transitions.

Method 3: Van't Hoff Fit

Fit the full two-state model with baseline corrections. Six parameters: F_N, F_U, m_N, m_U, T_m, ΔH_m. Most information-rich but requires clean data and well-defined baselines. ΔH_m from this fit is the van't Hoff enthalpy — compare to calorimetric ΔH from DSC to check two-state assumption.

Common Pitfalls

No visible transition

The protein is already unfolded, Tm is above instrument range, or the protein lacks a hydrophobic core. Run a known stable protein (e.g., lysozyme, Tm ≈ 72°C) as positive control.

High baseline fluorescence

The protein has exposed hydrophobic patches even when folded (common for antibodies, membrane proteins). Reduce SYPRO Orange concentration, or switch to nano-DSF.

Multiple transitions

Multi-domain proteins unfold domain-by-domain. This can be informative if transitions are separated enough to resolve, but overlapping domains can produce misleading apparent Tm values. Fit a sum of transitions only when the data support it.

DMSO artifacts

DMSO can shift Tm in a protein-dependent way. Must keep DMSO constant across all wells and normalize ΔTm to the matched DMSO control — not to the DMSO-free control.

Compound fluorescence

Some compounds fluoresce at the SYPRO Orange emission wavelength → artificial signal increase. Run a "compound + dye, no protein" control. Switch to nano-DSF if compound interference is widespread.

Aggregation-dominated curves

Irreversible aggregation produces broad, asymmetric transitions. Tm from a sigmoidal fit may be inaccurate. Report Tonset (temperature of initial fluorescence increase) instead.

Scan rate dependence

For irreversible unfolding, Tm depends on scan rate (faster scanning → higher apparent Tm). Standard: 0.5–1°C/min. Always report scan rate.

Confusing ΔTm with K_D

The single most common error in the TSA literature. ΔTm is a composite parameter that depends on KD, ΔH_binding, ΔH_unfolding, protein concentration, and ligand concentration. Do not report ΔTm as KD.

Strengths & Limitations

✅ Strengths	❌ Limitations
Runs on compatible real-time qPCR instruments	ΔTm ≠ KD (not a direct affinity measurement)
Very fast: 30–60 min per 96-well plate	SYPRO Orange binds detergent → unusable for membrane proteins (use nano-DSF)
Minimal protein: 1–5 µg per well	Proteins without hydrophobic core give no signal
HTS-compatible: 384-well and 1536-well possible	Cannot distinguish binding site or binding mode
Excellent for buffer/formulation screening	Irreversible unfolding is scan-rate dependent
Fragment screening at very high throughput	Compound fluorescence interference (dye-based)
No labeling, no immobilization	Multiple transitions complicate analysis
Nano-DSF: label-free and more compatible with detergents	Aggregation obscures thermodynamic parameters
Nano-DSF: backreflection/turbidity readout for aggregation onset (Tagg)	Weak binders may give ΔTm below detection limit
Extremely low cost per data point	Temperature-dependent KD complicates quantitative analysis

When NOT to Use TSA

❌ When you need K_D

TSA gives ΔTm, not KD. For affinity, use SPR, MST, ITC, or FP. Even the Schellman/isothermal analysis approach gives only approximate KD with significant assumptions.

❌ For membrane proteins with dye-based DSF

SYPRO Orange often binds detergent micelles → high background or compressed dynamic range. Use nano-DSF, CPM dye, or another detergent-compatible assay.

❌ For intrinsically disordered proteins

No cooperative folded core often means weak, broad, or non-classical thermal-shift signals. Nano-DSF may work if Trp/Tyr environments change during local folding, but interpretation is not a simple global Tm.

❌ When kinetic information is needed

No kon, koff, or residence time. TSA measures equilibrium stability — no kinetic information. Use SPR or BLI for kinetics.

❌ For covalent binders (without modification)

Standard TSA detects reversible stabilization. Covalent modifiers may stabilize or destabilize depending on modification site. Time-dependent incubation experiments can help, but interpretation is complex.

❌ When compound solubility is limiting

Fragment screening requires 1–10 mM compound. Many fragments precipitate at these concentrations → light scattering artifacts. Centrifuge plates before reading, or use nano-DSF with back-reflection.

TSA vs Related Techniques

🌡️ TSA / DSF

ΔTm (rank-ordering, not KD)
SYPRO Orange dye or nano-DSF (label-free)
qPCR machine or Prometheus
Fastest, cheapest, highest throughput
1–5 µg protein per well
No kinetics, approximate thermodynamics

🔄 CD Thermal Melt

Tm from secondary structure (optical)
Monitors backbone conformation (222 or 208 nm)
No dye → no dye artifacts
Lower throughput (1 sample at a time, ~1 h per melt)
More sample: 50–200 µg per experiment
Better for studying folding intermediates

🔥 DSC (Differential Scanning Calorimetry)

Direct calorimetric ΔH_cal and Tm
Gold standard for thermodynamic parameters
ΔH_cal / ΔH_vH ratio tests two-state assumption
Highest sample consumption (~0.5–1 mg)
Slowest (1–3 h per scan)
Use for detailed characterization of top hits from TSA

🧪 Chemical Denaturation

ΔG_unfolding at 25°C directly
Typically reversible (avoids irreversibility artifacts)
Can distinguish two-state from multi-state unfolding
Lower throughput: titration series per protein
Requires good spectroscopic probe
Use when you need ΔG at physiological temperature

TSA for Drug Discovery

🧩 Fragment Screening

384- or 1536-well formats can support very high-throughput fragment screens
Cocktail screening (5–10 fragments/well) for even higher throughput
Hit rate: typically 1–5% with ΔTm > 1°C
Orthogonal validation: SPR (binding kinetics), X-ray crystallography
Widely used primary or triage screen for FBDD

🎯 Hit-to-Lead Triaging

Dose-response ΔTm for confirmed hits → rank by apparent potency
Counter-screen against related proteins for selectivity
TSA ΔTm trends should correlate with orthogonal affinity measurements
Flag compounds with dose-dependent fluorescence (interference)

⚗️ Formulation Development

Screen 96 buffer conditions per plate
Variables: pH (4–9), NaCl (0–500 mM), additives (sucrose, trehalose, arginine)
Maximize Tm for long-term storage stability
Use DSF alongside orthogonal assays in formulation and comparability packages

Publication Checklist

Experimental Design

☐Protein concentration and source stated
☐SYPRO Orange concentration stated (or nano-DSF noted)
☐Buffer composition and pH stated
☐DMSO concentration stated and matched across wells
☐Scan rate stated (°C/min)
☐qPCR instrument and filter set stated
☐Number of replicates (≥ 3 technical replicates)
☐Positive control (known binder) included

Data Quality

☐Tm determination method stated (Boltzmann, derivative, van't Hoff)
☐Representative raw melt curves shown (not just ΔTm values)
☐ΔTm values with standard deviation / confidence interval
☐No-compound (DMSO-only) control Tm reported
☐Compound-only (no protein) control for fluorescence interference
☐For screening: Z′ factor reported (should be > 0.5). Z′ = 1 − 3(σ_p + σ_n) / |µ_p − µ_n|

Reporting

☐ΔTm reported — NOT called KD
☐Dose-response shown if available (not just single-dose)
☐Hits validated by orthogonal method (SPR, ITC, MST)
☐For nano-DSF: 350/330 ratio shown, Tonset and Tagg reported
☐Scan rate and heating direction stated
☐Irreversibility noted if applicable (no re-cooling data shown)

Instruments for DSF / Thermal Shift

Compatible real-time qPCR machines — Applied Biosystems QuantStudio, Bio-Rad CFX96/CFX384, Roche LightCycler 480, Eppendorf Mastercycler
NanoTemper Prometheus NT.48 — nano-DSF, 48 capillaries, intrinsic fluorescence + backreflection aggregation readout
NanoTemper Prometheus Panta — nano-DSF plus aggregation/particle-size readouts, 48 capillaries, automated loading
NanoTemper Tycho NT.6 — rapid nano-DSF, 6 capillaries, about 3 min per run, quality control focus
Unchained Labs UNit / UNCLE — multi-parameter: DSF, DLS, SLS in one instrument

Have SPR or BLI data?

Upload your raw files and get an automated kinetic analysis in minutes. We support Biacore, Octet, and other major formats.

Upload & Analyze