ThermoFAD, a Thermofluor®-adapted flavin ad hoc detection system for protein folding and ligand binding


F. Forneris and A. Mattevi, Dipartimento di Genetica e Microbiologia, Università di Pavia, Via Ferrata 1, 27100 Pavia, Italy
Fax: +39 0382 528496
Tel: +39 0382 985534


In living organisms, genes encoding proteins that contain flavins as a prosthetic group constitute approximately 2–3% of the total. The fluorescence of flavin cofactors in these proteins is a property that is widely employed for biochemical characterisation. Here, we present a modified Thermofluor® approach called ThermoFAD (Thermofluor®-adapted flavin ad hoc detection system), which simplifies identification of optimal purification and storage conditions as well as high-affinity ligands. In this technique, the flavin cofactor is used as an intrinsic probe to monitor protein folding and stability, taking advantage of the different fluorescent properties of flavin-containing proteins between the folded and denatured state. The main advantage of the method is that it allows a large amount of biochemical data to be obtained using very small amounts of protein sample and standard laboratory equipment. We have explored several cases that demonstrate the reliability and versatility of this technique when applied to globular flavoenzymes, membrane-anchored flavoproteins, and macromolecular complexes. The information gathered from ThermoFAD analysis can be very valuable for any biochemical and biophysical analysis, including crystallisation. The method is likely to be applicable to other classes of proteins that possess endogenous fluorescent cofactors and prosthetic groups.


lysine-specific histone demethylase 1


monoamine oxidase


Thermofluor®-adapted flavin ad hoc detection system


flavin-dependent monooxygenase

Identification of optimal purification and storage conditions is one the most critical investigations in the biochemical analysis of a protein. Challenging projects such as characterisation of macromolecular complexes, membrane proteins or large multidomain human proteins often do not provide the large amounts of sample required by protein biochemistry techniques, restricting the investigation to a very limited, sometimes not reproducible, set of information. In this respect, the Thermofluor® technique [1] (Fig. 1A) is an example of how it is possible to minimise the amounts of protein and time used for analysis of various parameters such as ligand stabilisation, pH effects, and storage conditions [2–4]. Thermofluor® determines the unfolding temperature of a protein through evaluation of the fluorescence of a solvatochromic dye such as 1-anilino-8-naphthalenesulfonate [5] or SYPRO Orange [6], which have a low fluorescence quantum yield in water and a high quantum yield when bound to the hydrophobic surface of denatured proteins (Fig. 1A). Over recent years, several reports have described successful use of the Thermofluor® technique for identification of the stabilising conditions of biochemically uncharacterised proteins [5–8], library screening of potential ligands for selected drug targets [9–11], or simple investigations of the behaviour of proteins under various conditions [12,13]. Although dedicated instruments are commercially available for Thermofluor® analysis [1], the experiment can be performed without any technical adaptation, using even the cheapest available real-time PCR apparatus [13,14]. The fluorescence signal is increased when the dye partitions into the hydrophobic patches of proteins that become solvent-exposed during the denaturation process. The presence of compounds interacting with the protein molecules at various levels, from solvation to covalent binding, alters the unfolding behaviour of the protein under analysis, and a shift in the unfolding temperature can be directly associated with a stabilisation or a destabilisation effect [4,11]. However, use of dyes that bind hydrophobic surfaces, such as SYPRO Orange, suffers from the limitation that the detergents used to solubilize membrane proteins interfere with the analysis, creating a hydrophobic environment due to micelle formation. This dye-detergent interaction does not allow correct measurement of the unfolding temperature of the sample, limiting the analysis to water-soluble proteins. For the same reason, Thermofluor® cannot be applied successfully to many proteins that expose hydrophobic patches to the solvent (e.g. proteins that interact in macromolecular complexes), because the dyes produce a fluorescence signal due to binding to these regions, masking the signal associated with protein unfolding.

Figure 1.

 (A) Schematic representation of the Thermofluor® binding assay. A solvatochromic dye (i.e. SYPRO Orange) is used as an indicator of protein unfolding. Binding of the dye to the unfolded protein results in a significant increase in its intrinsic fluorescence. (B) Schematic representation of ThermoFAD. In this case, the increase in fluorescence is generated by exposure of the flavin cofactor to the solvent upon protein unfolding. (C) Overview of fluorescence properties of flavins and comparison with RT-PCR instrumental parameters. Dashed line, flavin excitation spectrum; continuous line, flavin emission spectrum; red, wavelength range for RT-PCR fluorescence excitation; green, SYBR Green detection range; orange, SYPRO Orange detection range. Flavin fluorescence emission can be measured using the SYBR Green fluorescence filter on the RT-PCR instrument without any adaptation.

In both prokaryotic and eukaryotic organisms, genes encoding proteins that contain flavins as prosthetic group are estimated to constitute approximately 2–3% of the total. Enzymes that employ flavins for catalysis are involved in a multitude of processes, from drug metabolism to gene regulation [15]. Because of their spectroscopic features, flavoproteins form one of the most studied protein classes. In particular, the fluorescence of the flavin is an intrinsic property that is widely used for biochemical characterisation of flavoproteins. By comparing the emission and excitation ranges of the dyes typically used in Thermofluor® experiments, we noticed that flavins have fluorescence properties that fall in the same wavelength range. The conventional excitation wavelength used in RT-PCR instruments is 450–530 nm, whereas flavins show fluorescence excitation maxima at 373–375 and 445–450 nm (Fig. 1B,C) [16]. This broad shape of the flavin excitation spectrum makes the RT-PCR excitation wavelengths suitable for generating sufficient fluorescence intensity for detection. With regard to fluorescence emission, the highest intensity for flavins is at 535 nm [16]. Depending on the instrumental setup, RT-PCR instruments have various optical ranges for fluorescence detection, from fixed intervals to a completely customizable detection range [17]. However, we found that most RT-PCR systems, even the cheapest ones available on the market, can generally be used to excite flavins and measure their fluorescence signal without any specific adaptation. As the fluorescence of flavin cofactors in flavoproteins is usually quenched by the protein environment when the protein is properly folded [16], we realised that is possible to measure the unfolding temperature of a flavoprotein using Thermofluor® by monitoring the increase in cofactor fluorescence (Fig. 1B). This approach allows fast and reliable evaluation of many protein parameters using extremely low amounts of sample. Moreover, it is more versatile than conventional Thermofluor® because, by using intrinsic fluorescence instead of that of an external dye, it is not influenced by the noise generated by hydrophobic compounds present in solution or hydrophobic patches that may interact with the dyes used in Thermofluor®. We named this modified Thermofluor® approach ‘ThermoFAD’ (Thermofluor®-adapted flavin ad hoc detection system).


The ThermoFAD technique

A ThermoFAD analysis requires only 20 μL of protein sample, in a concentration range from 0.3 to 4.0 mg·mL−1, and an RT-PCR instrument. The whole experiment takes < 2 h and allows evaluation of 1–384 samples at the same time (depending on the set-up of the PCR instrument). In a typical experiment, 1–2 μL of a concentrated sample are added together with the buffers and ligands for analysis directly into the wells of the RT-PCR instrument. Next, a temperature gradient is applied, starting from 15–20 °C and increasing to 90 °C, measuring the fluorescence signal every 0.5 min. As in a standard RT-PCR Thermofluor® experiment, a sigmoidal curve (thermogram) is obtained by plotting the fluorescence intensity against the temperature. The unfolding temperature is then determined as the maximum of the derivative of this sigmoidal curve (Figs 2 and 3) [1]. By comparing various thermograms for the same protein under various conditions, it is possible to evaluate which compounds stabilise (or destabilise) the sample under analysis and to screen many conditions with a minimum consumption of protein [5].

Figure 2.

 Comparison between Thermofluor® and ThermoFAD for various flavoproteins. The selected flavoproteins differ with respect to the type of flavin cofactor, flavin linkage to the protein, and source organism of the protein (for details see Table 1). Thermal stability curves are plotted against normalised fluorescence signal. Green lines, Thermofluor® experiments using SYPRO Orange as fluorescent dye; red lines, ThermoFAD experiments measured without addition of any dye. The detector filter of the RT-PCR instrument for ThermoFAD is the one that is commonly used for SYBR Green dye (fluorescence emission of 523–543 nm; see Fig. 1C).

Figure 3.

 (A) Evaluation of FMO stability using ThermoFAD against various buffers at various pH values. (B) ThermoFAD comparison of LSD1 stability with (red) and without (green) addition of the protein CoREST. The Tm shift corresponds to formation of a heterodimeric complex between the two proteins. The Thermofluor® profile of isolated CoREST is shown in blue; in this case it is not possible to calculate a Tm value because of the many exposed hydrophobic patches of CoREST that bind to the dye before complete unfolding of the protein. (C) ThermoFAD profiling of LSD1/CoREST stability towards known inhibitor peptides. All data are in good agreement with the biochemical analysis [23]. In particular, the Lys4Met (K4M) peptide shows the highest stabilising effect, in agreement with the fact that it is the peptide that allowed us to solve the crystal structure of the LSD1/CoREST complex with a bound peptide substrate analogue.

In order to validate our ThermoFAD technique, we have chosen a set of flavoproteins with various features in terms of biological activity, size, and type of interaction (covalent and non-covalent) with the flavin cofactor (Table 1). As an indication of the efficiency and sensitivity of our approach, we compared the results obtained with ThermoFAD with conventional Thermofluor® measurements obtained using SYPRO Orange as the fluorescent probe for denaturation (Fig. 2). The results are in perfect agreement for the whole set of flavoproteins under analysis (Table 1). The sensitivity and specificity of ThermoFAD make the technique extremely versatile, in that it allows evaluation of the stability of a flavoprotein even in partially purified samples (data not shown), which is impossible to detect using dyes that bind nonspecifically to all the hydrophobic patches present in solution. Here, we report on the application of the modified Thermofluor® approach to a few of our investigated flavoproteins (Table 1 and Fig. 2) with the intention of demonstrating the advantages offered by the ThermoFAD technique.

Table 1.   Comparison of unfolding temperature using Thermofluor® and ThermoFAD for various flavoproteins. ND, not determined.
ProteinOrganismBound flavinReferenceProtein concentration (mg·mL−1)Tm (°C)
Lysine-specific demethylase + CoREST complexMammal (Homo sapiens)Non-covalent FAD[23]
Polyamine oxidasePlant (Zea mays)Non-covalent FAD[25]0.650.050.2
l-Galactono-γ-lactone dehydrogenasePlant (Arabidopsis thaliana)Non-covalent FAD[26]1.358.258.6
Flavin-dependent monooxygenaseBacterial (Methylophaga sp.)Non-covalent FAD[18]
Monoamine oxidase BMammal (Homo sapiens) Covalent (Cys) FAD[27]1.0ND51.2
Alditol oxidaseBacterial [Streptomyces coelicolor A3(2)]Covalent (His) FAD[28]1.549.749.4
Cytokinin dehydrogenasePlant (Zea mays)Covalent (His) FAD[29]1.059.960.3
Vanillyl-alcohol oxidaseFungus; (Penicillium simplicissimum)Covalent (His) FAD[30]

Comparison of the stability of a soluble and globular flavoenzyme (FMO) in the presence of various ligands

We tested a number of conditions for optimal stabilisation of a flavin-containing monooxygenase (FMO) from Methilophaga sp. strain SK1. FMOs are involved in the metabolism of several drugs, catalysing the oxygenation of many nitrogen-, sulphur-, phosphorus- and selenium-containing nucleophilic compounds using molecular oxygen and NADPH as substrates [18]. Using ThermoFAD, we compared the stability of FMO in various buffers and evaluated the effect of addition of NADP(H) analogues on protein stability. Our buffer screening led to identification of optimal stabilisation conditions for FMO that correspond to the buffer that was successfully used for crystallisation of this flavoprotein [18] (Fig. 3A). Moreover, the ThermoFAD analysis allowed us to identify NADP analogues with higher affinity to FMO compared to NADP:3-acetylpiridine ADP, thioNADP and nicotinic acid ADP. These compounds were then tested as FMO crystallisation additives, leading to high-quality crystals, with a significant increase in the diffraction quality and resolution of the data (F. Forneris and A. Mattevi, unpublished results).

ThermoFAD on a membrane-anchored flavoenzyme in the presence of detergents

When working with membrane proteins, it is necessary to use detergents after membrane extraction throughout the purification and characterisation process. The choice of detergent is the most critical parameter in obtaining a stable and active protein suitable for biochemical and structural characterisation. For this reason, effective detergent screening methods are required (see [19] for a recent development in this area). Thermofluor® is an excellent candidate for this type of analysis, but suffers from the limitation that the fluorescent dyes used to determine the protein unfolding temperature interact with the detergent lipophilic moiety. This limitation makes the analysis difficult, if not impossible [20]. However, ThermoFAD allowed unfolding temperature analysis of a membrane-anchored flavoprotein to be performed in the presence of detergents, because the flavin cofactor fluorescence is not influenced by these amphipathic molecules. As a test case, we used human monoamine oxidase B, a membrane-bound flavoenzyme that catalyses the oxidation of arylalkylamine neurotransmitters and bears a FAD cofactor covalently attached to a cysteine residue [21]. We performed both Thermofluor® (using SYPRO Orange as a dye) and ThermoFAD experiments on the same sample in order to compare the two techniques. The Thermofluor® experiment did not produce a sigmoidal curve, most likely because of interaction of SYPRO Orange with the detergent and/or the hydrophobic membrane-binding region of the enzyme. On the other hand, ThermoFAD produced a clear result with a nice sigmoidal curve indicating an unfolding temperature of 51.2 °C (Fig. 2). The significance of this result was further verified by circular dichroism spectropolarimetry. By means of this technique, we measured a value for the unfolding temperature (57 °C) that is slightly higher than that measured by ThermoFAD, probably reflecting the inherent differences between the two methodologies. ThermoFAD senses the exposure of flavin to water, which is likely to be an earlier event in the denaturation process than the loss of secondary structures, as probed by circular dichroism. Our study of human monoamine oxidase B shows that ThermoFAD can be efficiently used in the case of flavoproteins that require detergents for stabilisation or that contain hydrophobic patches on their surface.

Evaluation of in vitro reconstitution of a protein complex using ThermoFAD

A more complicated case is an investigation conducted on the human flavin-dependent histone demethylase LSD1. This flavoenzyme catalyses removal of a methyl group from a protein substrate (histone H3) with a highly specific substrate specificity (Lys4). LSD1 is a partially non-globular, multidomain protein that is known to interact with a co-repressor protein named corepressor of the neural receptor REST (CoREST). LSD1 and CoREST assemble to generate a heterodimeric sub-complex that is part of several nuclear multiprotein complexes [22]. Using ThermoFAD, we were able to measure the stabilising effect induced by association of CoREST with purified LSD1 (Fig. 3B). Binding of CoREST to LSD1 shifts the unfolding temperature by 4 °C, consistent with a tight association between the two proteins. Thus, the experiment allowed us to quickly establish using a very limited amount of protein that the complex could be reconstituted in vitro. Moreover, we confirmed that various histone H3 peptides bind tightly to LSD1, in perfect agreement with biochemical enzymatic assays [23]. Importantly, the increases in protein stability are proportional to the inhibitory power of the analysed peptides (Fig. 3C). Especially interesting is the finding that the histone H3 peptide with the Lys4Met mutation has the highest stabilising effect. This peptide is a tight nanomolar inhibitor, which was successfully used for crystal structure determination of the LSD1/CoREST/histone peptide ternary complex [24].


Our method shows that it is possible to exploit the intrinsic fluorescence of flavin cofactor to determine the unfolding temperature of flavoproteins, instead of using the fluorescent dyes commonly used in Thermofluor® experiments. This approach simplifies the screening and identification of optimal conditions for protein stability, storage and ligand binding. In addition, this technique does not require any customised procedure or specific chemical compound, and can also be used in the presence of compounds that are known to interfere significantly with the dyes used in the conventional Thermofluor® approach, such as detergents or contaminants. We have provided some examples of the versatility of this technique, which can be used with proteins with covalently and noncovalently bound flavin cofactors to identify stabilising agents, high-affinity ligands, protein complex formation, and other factors that can affect protein stability. This information is obviously very valuable for any biochemical and biophysical analysis, including crystallisation. In all cases, the experiments were performed in just a few hours using standard laboratory equipment with minimal sample consumption. As flavoproteins are among the most widely studied protein classes because of their abundance, variety and biological importance, we believe that this fast, cheap and reliable method will be of great help for the many groups that study new and uncharacterised flavoproteins. Moreover, it is likely to be applicable to other classes of proteins that possess endogenous fluorescent cofactors and prosthetic groups.

Experimental procedures

Protein samples

All flavoproteins used for our analysis were expressed and purified as described in the original papers reporting their biochemical and structural characterisation (Table 1). Their purity was checked by SDS–PAGE analysis, and protein concentration was evaluated by measuring the UV/vis absorbance of the bound flavin cofactor using published extinction coefficients.

ThermoFAD experimental setup

Experiments were performed using a MiniOpticon real-time PCR detection system, using 48-well RT-PCR plates (Bio-Rad Laboratories, Hercules, CA, USA). Measurements were performed using an excitation wavelength range between 470 and 500 nm and a SYBR Green fluorescence emission filter (523–543 nm), which falls within the same fluorescence range as the isoalloxazine ring of FAD or flavin mononucleotide (470–570 nm) (Fig. 1C). The flavoprotein concentration required for optimal signal-to-noise ratio was initially evaluated using LSD1 as a benchmark. Unfolding curves were generated using a temperature gradient from 20 to 90 °C, performing a fluorescence measurement after every 0.5 °C increase after a 10-s delay for signal stabilisation. All experiments were performed at least three times, and the reported Tm values are based on the mean values determined from the peaks of the derivatives of the experimental data. In a typical experiment, 1–2 μL of a concentrated protein were mixed together with the ligands for analysis directly into the wells of the RT-PCR instrument and diluted with reaction buffer (50 mm KPi, pH 7.5) to a final volume of 20 μL. The best concentrations for ThermoFAD analysis were between 0.5 and 4 mg·mL−1, and all subsequent experiments were carried out using protein concentrations in this range.

Evaluation of the reliability of ThermoFAD versus Thermofluor® for various flavoproteins

To compare the results of the ThermoFAD analysis with conventional Thermofluor®, we performed experiments in parallel with the same amounts of flavoproteins, with and without the addition of 3 μL of 5000× SYPRO Orange (Sigma-Aldrich, St Louis, MO, USA). The experimental setup, gradients and methods were identical in the ThermoFAD and Thermofluor® analyses. Detection was performed using the SYPRO Orange and SYBR Green fluorescence filters for both techniques to evaluate the interference possibly caused by superposition of the flavin fluorescence on that of the SYPRO Orange. No interference was detected (data not shown).

Determination of stabilisation conditions for FMO

FMO was concentrated using an Amicon concentrator (Millipore Corp., Billerica, MA, USA) with a 30 kDa cutoff to a final concentration of 20 mg·mL−1. A set of 15 buffers at 50 mm concentration in the pH range 4.2–10.6 was prepared in RT-PCR plates, and 2 μL of flavoenzyme were added to each well (final protein concentration of 2.0 mg·mL−1). Buffers that showed a significant stabilisation effect are reported in Fig. 3A.

Determination of the unfolding temperature of human monoamine oxidase B

Human monoamine oxidase B, stored at 3 mg·mL−1 in 50 mm KPi pH 7.0 supplemented with 0.8% w/v octylglucoside, was diluted in the same buffer to a final concentration of 1 mg·mL−1 and used for thermal unfolding assays. The unfolding temperature was also measured by circular dichroism spectropolarimetry. For this purpose, we used a Jasco J-710 spectropolarimeter (Jasco Europe, Cremella, Italy) equipped with a Neslab RT-11 programmable water bath (Thermo Fisher Scientific, Waltham, MA, USA) and a 1 mm path-length cuvette. Thermal denaturation was followed by continuous measurements of ellipticity at 222 nm in the temperature range 25–70 °C with a constant heating rate of 1 °C·min−1.

LSD1/CoREST reconstitution and inhibition assays

Human LSD1, 8 mg·mL−1 in 50 mm KPi buffer supplemented with 5% v/v glycerol pH 7.2, was diluted with the same buffer to a final concentration of 1 mg·mL−1. Experiments were performed using the LSD1 alone or supplied with human CoREST in stoichometric amounts to determine the Tm increase associated with formation of the heterodimeric protein complex. For inhibition assays, a tandem-affinity purified LSD1/CoREST complex [24] was used instead of LSD1 alone for better comparison with previously published biochemical data [23]. The complex was used at a final concentration of 1 mg·mL−1, and 3 μL of 2 mm histone peptide inhibitors were added to each well.

Copyright notice

The Thermofluor® assay was developed by 3-Dimensional Pharmaceuticals Inc., which is now part of Johnson & Johnson Pharmaceutical Research & Development (Raritan, NJ, USA). ‘Thermofluor®’ is a trademark registered in the USA and certain other countries.


Financial support from the Italian Ministry of Science (PRIN06 and FIRB programmes), the Fondazione Cariplo, the Italian Association for Cancer Research, and the American Chemical Society Petroleum Research Fund (46271-C4) is gratefully acknowledged. We thank Drs Dale E. Edmondson (Emory University, Atlanta, GA, USA), Claudia Binda (University of Pavia, Italy), Willem J. van Berkel (University of Wageningen, the Netherlands) and Marco W. Fraaije (University of Groningen, the Netherlands) for providing us with protein material and helpful advice.