A beginner’s guide to differential scanning fluorimetry

Differential Scanning Fluorimetry (DSF) is used to measure the relative stability of proteins in vitro. This article provides an overview of DSF, alongside potential applications of the technology and a troubleshooting guide for assay optimization.

DSF, also called thermal shift assay (TSA), is a biophysical technique used to determine the apparent melting temperature (T_ma) of a protein. In a typical DSF experiment, a purified protein is mixed with a dye and the solution is heated in a polymerase chain reaction (PCR) instrument. A key feature of the dye is that it must have low fluorescence in aqueous solution, but be brightly fluorescent when in a hydrophobic environment. As the heated protein unfolds, hydrophobic amino acid sidechains that were previously restricted to the protein’s interior become exposed, creating favorable interactions with the fluorescent dye. Accordingly, a plot of temperature vs relative fluorescence units (RFUs) creates a characteristic DSF curve, with low initial fluorescence and then an increase with temperature (Figure 1). In some cases (but not all), the fluorescence will then decay slightly at higher temperatures, probably due to protein aggregation and partial masking of potential dye-binding sites. Regardless, the T_ma value is calculated as the half-maximal temperature of the protein unfolding curve. This value is an indication of the protein’s relative stability; a well-folded, stable protein will have a high T_ma value, while an unstable protein will have a low T_ma value and require less heat to denature.

Knowing the stability of a protein can be very informative. For example, this simple measurement can be exploited to measure the binding of proteins to ligands, such as inhibitors or cofactors (Figure 2a). In this case, the putative ligand is added to the protein and its binding is indicated by a rightward shift in the T_ma value (Figure 2b). The difference between the T_ma values of the ligand-bound sample and the control sample is reported as the ΔT_ma. In this way, a series of simple DSF measurements can be used to monitor the binding of a protein to a wide range of ligands, including small molecules, ions, metabolites, substrates, or peptides. Moreover, under the proper set of conditions, the value of the ΔT_ma can suggest the affinity of the interaction.

DSF has also been used to explore the effects of mutations on the relative stability of a protein (Figure 2c). In this use-case, the T_ma of the wild-type protein is compared to the T_ma values of mutants (e.g., variants containing point mutations). If the mutant is more stable than the wild type, then the ΔT_ma will be positive, while the ΔT_ma will be negative for a destabilizing mutation. Thus, such experiments reveal the relationships between a protein’s sequence and its stability. This information can be biomedically important because disease-associated mutations sometimes cause the protein to become unstable.

In another popular use of DSF, the impact of buffers and/or buffer components on protein stability can be explored (Figure 2d). For example, screening a protein in different buffers might reveal that a specific buffer component, such as divalent cations, strongly enhances thermal stability. For this reason, structural biologists often use DSF to select buffer conditions that maximize a protein’s stability, increasing the likelihood of acquiring high-quality structural information.

An emerging use of DSF is to study protein–protein interactions (PPIs) (Figure 2e). In the simplest case, this approach can be used to monitor homooligomerization of a protein because the complex will be more stable than the isolated monomer. In some cases, DSF might also be used to study interactions between two different proteins (e.g., heterodimers); however, this application requires that the individual T_ma values are sufficiently distinct from each other (see Further Reading for some examples).

Regardless of how DSF is being used, it is important to mathematically fit the fluorescence vs temperature curve to determine an accurate T_ma value. The simplest, and most commonly used, approach is to fit the curve to a first derivative. This method can be readily performed in most commercial software packages and works well for data that approximates the idealized case (as in Figure 1). However, many “real-life” DSF curves will deviate from a well-formed, sigmoid shape. For example, there might be high initial fluorescence, multiple transitions, or a temperature-dependent decay in fluorescence. These nonidealized features are often “real” behaviors of the system and should not be ignored. Rather, specially designed software tools, such as DSFworld (https://gestwickilab.shinyapps.io/dsfworld/), have been introduced to provide an expanded set of fitting algorithms (see Further Reading). For beginners, it is important to critically evaluate the quality of your data fitting to ensure that the T_ma values you calculate (and report) are accurate and that they faithfully reflect the raw data.

DSF is not the only way to measure the stability of a protein and platforms such as circular dichroism (CD) and differential scanning calorimetry (DSC) are considered to be the “gold standard”. So why use DSF? Compared to other techniques, DSF can be performed in high throughput. In a typical experiment, only ~10 to 20 µL of protein (~2 to 10 µM) is needed such that the experiment can be performed in 384-well PCR microtiter plates. Moreover, a DSF experiment only takes around 60 minutes to complete and it can be performed in a standard tabletop PCR instrument. These features allow testing of hundreds or thousands of perturbations in most research laboratory settings, including those used for undergraduate research. However, it is important to note that DSF has limits. For example, thermal denaturation often introduces irreversible, kinetically driven aggregation steps (e.g., note the fluorescence decrease at high temperatures in Figure 1). Thus, it is often incorrect to assume that a DSF experiment is at thermodynamic equilibrium. As always, it is essential to understand the strengths and weaknesses of each biophysical method, including DSF.

To use DSF rigorously, one must be aware of artifacts that are commonly encountered. Some of these artifacts arise from the microtiter plates themselves; for example, plates might warp during heating, potentially creating inaccurate fluorescence patterns. Likewise, microtiter plates from some vendors might have intrinsic fluorescence that partially masks the dye signal. We recommend testing each lot of plates for compatibility prior to starting an experimental campaign.

In our experience, the most common issues with DSF are caused by aberrant interactions of the dye with a nonprotein component of the system (e.g., buffer or ligand). One direct way to test for this possibility is to include a control in which the protein is left out of the sample (termed an “Everything But the Protein” or “EBP” control). More broadly, one should always include a complete series of controls that systematically isolate and test for potential artifacts (i.e., buffer only, buffer + dye, buffer + protein + dye, etc.). Ideally, the controls, including the EBP control, should produce low baseline fluorescence (as in Figure 1, dotted grey line). However, in some cases, the dye interacts with some other component of the sample, yielding a curve that deceptively resembles a genuine protein unfolding profile (Figure 3a; blue line). For more information, detailed DSF protocols have been reported (see Further Reading), providing users with proscribed ways of troubleshooting and avoiding these artifacts.

A related category of artifact is caused by various other types of dye incompatibility. For example, the dye might bind to the protein’s folded state or fail to bind to the unfolded state(s) (Figure 3b). In either case, the resulting temperature vs fluorescence curve will be challenging to fit and it will likely not produce an accurate T_ma value. One can sometimes avoid these issues by reducing the concentration of the dye. In cases where that approach does not fix the problem, newer techniques are also starting to address this issue by introducing alternative dyes (see below).

Finally, some proteins routinely produce poor DSF curves. Although the molecular origins of this incompatibility are not always clear, we have noted that switching the heating protocol to an “up-down” mode can sometimes improve the data quality. Briefly, standard DSF protocols involve continuous heating (e.g., temperature is increased by a set increment per unit of time; 1 °C/min). The alternative is to use iterative heating cycles (termed “up-down” mode), in which the sample is progressively heated but allowed to cool between each increment. Using this heating mode can sometimes produce vastly superior DSF curves for the same protein, even under the same buffer conditions. Further details on using these approaches are found in the Further Reading.

Recent advances have expanded the utility and scope of DSF applications. For example, new instruments have been developed to conduct nanodifferential scanning fluorimetry (nanoDSF) experiments. In this approach, the dye is left out of the system and, instead, thermal denaturation is monitored via the fluorescence of the protein’s own tryptophan residues. The conceptual basis for nanoDSF is that tryptophan residues often undergo a change in their chemical environment during protein unfolding, which can then be measured at 350 and 330 nm. This method has advantages for some systems, especially if there are dye incompatibilities (see Troubleshooting DSF Assays), but the relatively weak fluorescence intensity of tryptophan necessitates the use of specialized instruments and higher protein concentrations. Moreover, some proteins lack the requisite residues in the right positions. Still, nanoDSF provides a useful supplement to the field and an expansion of the technology.

Another exciting advance has been to replace the traditional DSF dye, SYPRO Orange (SO), with other dyes. As mentioned above, a dye can sometimes bind to the protein’s folded state or fail to bind to the unfolded state (see Figure 3b). In addition, SO is sometimes incompatible with buffers or ligands (see Figure 3a). Accordingly, alternative dyes, such as bis-ANS, Proteostat, or Glomelt, have been used as replacements. Our group recently reported on a library of >300 solvatochromatic dyes, termed the Aurora collection, which we used to identify compatible dyes for >90% of the tested proteins (see Further Reading). In short, the dye has become a variable in a DSF assay, which can now be replaced during assay optimization.

The availability of both next-generation dyes and more powerful fitting software have also inspired the use of DSF to answer increasingly sophisticated questions. For example, some Aurora dyes allow discrimination of two clear and discrete melting transitions (commonly termed T_ma1 and T_ma2) in multidomain proteins. In the systems that have been explored thus far, these transitions seem to correspond to the thermal unfolding of the individual domains in the target protein. Thus, in theory, this feature allows testing of the impact of mutations or ligands on the relative stability of one domain vs others. The expanded uses of DSF have led to the coining of the term “paDSF” or protein-adaptive differential scanning fluorimetry to indicate that the technique is now able to be used in new ways that are tuned (or “adapted”) to the specific question or protein system in unprecedented ways. It seems likely that further improvements in software, dyes and instruments will continue to expand the scope of DSF experiments.

Our work on DSF has been funded by the Tau Consortium and the US National Institutes of Health (R01GM141299).

Annemarie Charvat earned a B.S. in Chemistry from the University of San Francisco, where she performed undergraduate research on bioinorganic chemistry. She then joined the Gestwicki Laboratory, where she is working to improve DSF technology. Email: [email protected]. Twitter: @anniecharvat

Kayleigh Mason-Chalmers earned a B.A. in Chemistry from Vassar College, performing undergraduate research in protein biochemistry and neurodegeneration. In 2024, she joined the Gestwicki Laboratory, where she works to expand the scope of DSF methods. Email: [email protected].

Jason E. Gestwicki earned a Ph.D. in Biochemistry from the University of Wisconsin-Madison and completed postdoctoral studies at Stanford University. He is currently a Professor in the Department of Pharmaceutical Chemistry at University of California San Francisco (UCSF). His laboratory uses chemical biology approaches to explore how protein stability and protein homeostasis impact human health and disease. Email: [email protected]. Twitter: @GestwickiLab