263

Wayne F. Anderson (ed.), Structural Genomics and Drug Discovery: Methods and Protocols, Methods in Molecular Biology, vol. 1140, DOI 10.1007/978-1-4939-0354-2_20, © Springer Science+Business Media New York 2014

Chapter 20

Ligand Screening Using Fluorescence Thermal Shift Analysis (FTS)

Chi-Hao Luan , Samuel H. Light , Sara F. Dunne , and Wayne F. Anderson

Abstract

The fl uorescence thermal shift (FTS) method is a biophysical technique that can improve productivity in a structural genomics pipeline and provide a fast and easy platform for identifying ligands in protein function or drug discovery screening. The technique has gained widespread popularity in recent years due to its broad-scale applicability, throughput, and functional relevance. FTS is based on the principle that a protein unfolds at a critical temperature that depends upon its intrinsic stability. A probe that will fl uoresce when bound to hydrophobic surfaces is used to monitor protein unfolding as temperature is increased. In this manner, conditions or small molecules that affect the thermal stability of a protein can be identifi ed. Herein, principles, protocols, data analysis, and special considerations of FTS screening as performed for the Center for Structural Genomics of Infectious Diseases (CSGID) pipeline are described in detail. The CSGID FTS screen is designed as a high-throughput 384-well assay to be performed on a robotic platform; however, all protocols can be adapted to a 96-well format that can be assembled manually. Data analysis can be performed using a simple curve fi tting of the fl uorescent signal using a Boltzmann or double Boltzmann equation. A case study of 100 proteins screened against Emerald Biosystem’s ADDit™ library is included as discussion.

Key words Protein ligand , Protein folding , Thermal shift , High-throughput screening , Drug discovery

1 Introduction

When integrated into a structural genomics pipeline, complimentary biochemical and biophysical techniques can enhance protein struc-ture determination productivity and provide a platform for gener-ating novel insights into protein function. Since the fl uorescence thermal shift (FTS) method was fi rst described in 2001, the tech-nique has gained widespread popularity and been effectively applied to address a variety of drug screening and general biophysical ques-tions [ 1 ]. Providing an effi cient medium- to high- throughput method for monitoring protein thermal denaturation across multiple conditions, FTS has increasingly found a place within structural genomics operations [ 2 – 4 ]. Several features suit FTS for

264

the structural genomics context and explain its growing popularity within the fi eld:

1. FTS has broad-scale applicability. As the need for customization is limited, FTS is suitable for handling the hundreds of func-tionally and structurally diverse proteins found within a typical structural genomics pipeline.

2. FTS allows for the characterization of protein stability a cross a range of conditions. Finding a stable condition can be a critical for crystallization and thus data provided by FTS can help maximize crystallization success.

3. FTS can identify biological ligands. Functionally uncharacter-ized proteins frequently constitute a sizable fraction of targets within the structural genomics pipeline. Screening simple libraries by FTS presents a cost-effective method for identifying unknown biological ligands, the identifi cation of which contrib-utes to the overall scientifi c output and provides co-crystallization opportunities that increase the likelihood of crystallographic success.

4. FTS can serve as a primary screening method for protein tar-gets that lack a suitable functional or binding assay to identify leads for drug discovery.

In this chapter, the basic principles behind FTS are reviewed and methodological and analytical considerations pertinent to the application of FTS in the structural genomics context and in drug discovery are discussed. For a more thorough review of the prin-ciples behind FTS see ref. 5 .

The FTS technique relies on the principle that proteins unfold at a critical temperature. At lower temperatures, proteins adopt a native state, which generally consists of compact and predictable tertiary structure. At higher temperatures, proteins denature, losing sec-ondary and tertiary elements and forming a molten globule or aggregates. Thermal stability is an intrinsic property that results from protein sequence and determines the temperature at which thermal denaturation occurs. Thermal stability can be measured by monitoring protein denaturation while incrementally raising sam-ple temperature and is conveniently quantifi ed as the temperature that marks the midpoint of thermal denaturation ( T m ).

FTS provides a convenient method for measuring protein T m . The method takes advantage of changes in intrinsic (i.e., from protein tryptophans) or extrinsic (e.g., from dyes, such as anilinon-aphthalene sulfonate or SYPRO Orange) fl orescence that accom-pany protein denaturation. As they reliably provide a strong signal, extrinsic dyes are more commonly used in current practice. The use of extrinsic dyes relies upon a simple principle. Driven primarily by the hydrophobic effect, proteins tend to adopt three- dimensional

1.1 Principles of FTS

Chi-Hao Luan et al.

265

structures with hydrophobic residues disproportionately buried at the core, resulting in a more polar external layer and a more hydro-phobic internal core. In the protein’s native state the internal hydrophobic core is sheltered by the external layer and inaccessible to the bulk solvent. Denaturation disrupts the tertiary structure of the protein and exposes core residues, increasing the number of solvent accessible hydrophobic residues. Effective FTS dyes sub-stantially increase their quantum yield when interacting with hydrophobic residues, ensuring that a measurable fl orescent signal is produced when denaturation results in the exposure of hydro-phobic core components.

It is a well-established thermodynamic phenomenon that ligand binding affects protein thermal stability. This connection between binding and thermal stability allows ligands to be identi-fi ed on the basis of their effect on protein T m . Prior to the advent of FTS, it would have been impractical to screen the effect of ligands on protein stability on all but the smallest scales. However, because the necessary components of an FTS experiment are quite basic (requiring only purifi ed protein, appropriate dye, temperature- controlled apparatus, and a fl uorescence detector) medium- to high-throughput FTS screens are achievable.

FTS has found a number of disparate applications within the protein sciences. The technique has been used as a quality control measure for protein production, as a means to determine the impact of point mutations on protein folding and stability [ 6 , 7 ], as a method to screen storage buffers [ 8 – 11 ], as a way to examine the effects of urea and other denaturants or glycerol and other protectants [ 12 , 13 ], and as a technique to investigate ligand binding [ 1 , 14 , 15 ]. This chapter focuses on FTS screening applications for the identi-fi cation of protein ligands and suitable crystallization conditions.

In the structural genomics context, studied proteins are often well- expressed and usually screening of only a limited number of conditions is called for (<10,000 and often <1,000 conditions). For this reason, sample consumption is generally not a major concern. Therefore, FTS can be applied routinely as a protein char-acterization measure, regardless of protein function and the exis-tence, or lack of, alternative functional assays.

FTS has two crystallization applications, to identify solvent/solute conditions and small molecule ligands. The former affects the protein through solvent effects or solute-mediated solvent effects. Ligand screening, on the other hand, is meant to identify the specifi c interaction of a small molecule with the protein. In this regard, ligand screening is similar to the task of drug discovery; however it usually does not require as high a binding affi nity as is necessary for a therapeutic agent. Due to differences in the confi guration and the screening concentrations of the two types of

1.2 Applications

1.3 Unique Concerns for Crystallography Application

Fluorescence Thermal Shift Screening

266

libraries, the protein preparation and buffer volumes differ between the two tactics. For solvent and solute conditions the screening concentration is in the 1–100 mM range, therefore a 1× library in buffer is prepared for direct use in the FTS screen, and a small amount of high concentration stock protein is added to the conditions. However, for ligand screens, small molecules at concentrations in the 5–50 μM range are suffi cient for 1:1 or similar stoichiometry.

Within drug discovery applications, FTS can be used either as a primary or as a secondary screen. The amount of protein required for the many conditions contained within a typical primary screen can be problematic. Before choosing FTS as a primary screening method one should consider:

1. Is an alternative functional assay available? 2. What is the total cost of a functional assay (substrate, cofactors,

detection methods, HTS availability, etc.)? 3. What is the total cost of an FTS assay?

If, after weighing these factors, the decision is made to proceed with an FTS primary screen, measures can be taken to minimize sample requirements. For example, miniaturization of assay volume and utilization of a pooled screening format (i.e., screening two or more compounds per well) can reduce the material cost.

While the substantial protein requirement can make FTS a suboptimal primary screening option, FTS frequently provides an excellent resource for secondary screening. Assuming a limited number of hits are identifi ed in the primary screen, minimal resources are required for secondary FTS screening. As a secondary screen, FTS can be conducted in a timely fashion to help weed- out false positives, reducing the number of compounds that needed to be screened in expensive and diffi cult in vivo functional assays.

2 Materials and Instrumentation

FTS assay detection utilizes environment sensitive dyes that fluoresce in hydrophobic environments. Three examples of dyes that can serve as reporters for protein unfolding are: (1) Nile red (Ex 585 nm, Em 665 nm), (2) SYPRO Orange (Ex 492 nm, Em 575 nm), and (3) bis-ANS (Ex 350 nm, Em 492 nm) or 1,8-ANS (Ex 350 nm, Em 492 nm). These represent fl uorophores in the red, green, and blue spectrum, respectively. Although ANS-based molecules have been used widely in protein investigations and in the early FTS experiments, SYPRO Orange has become the pre-ferred probe molecule for FTS in recent years. One of the main reasons that SYPRO Orange has gained popularity is that most commercially available qPCR instruments only support Ex/Em

1.4 Unique Concerns for Drug Discovery Applications

2.1 Dye

Chi-Hao Luan et al.

267

fi lters above 400 nm due to their light source (tungsten lamp or LED). Therefore, the lower wavelength bis-ANS and 1,8-ANS dyes are incompatible with these machines. SYPRO Orange also has a higher quantum yield increase when in hydrophobic environ-ment, providing an easily measured signal to noise ratio. Nile Red, however, remains a viable option for FTS. Particularly for drug discovery screens, fl uorescent compounds, which are most pro-nounced in the green range, can interfere with the dye signal. Because typical libraries produce less interference within its excitation/emission spectra, Nile Red provides an alternative for screening purposes to, in applicable cases, delineate degenerated signal.

For the purposes of this protocol, 5,000× SYPRO Orange in DMSO obtained from a commercial vendor (Life Technologies, Bio-Rad, or Sigma-Aldrich) is added either by directly mixing with buffer/protein stock solution or by a nanoliter liquid handling robot.

HEPES is the most commonly used buffer in FTS experiments. The advantage of HEPES for FTS lies in its pH stability with respect to temperature change. Dependent upon the buffering strength requirement, either higher or lower concentration, e.g., 100 or 20 mM HEPES can be used. A prescreen test can be helpful for selecting a buffer suitable for large-scale screening. NaCl is usually included in the buffer at 150 mM concentration.

Arraying a number of disparate small libraries in a high-throughput convenient format is preferable to relying upon a single master library, as this provides the fl exibility to customize the screening strategy to best meet the experimental objective. For a given experiment, appropriate library selection depends on the goals of the screen.

To identify biological ligands for proteins of unknown func-tion, it is useful to screen a library that contains common biological metabolites. As of June 2013, there were 14,747 ligands in PDB structures. Screening all PDB ligands would be impractical due to the prohibitive expense and would be unnecessary since the major-ity (10,092 out of 14,747) appear in a single structure. The top 21 are the 20 amino acids plus water, which appear in 71,614–86,016 structures. The next population group consists of 28 ligands that are present in1,009–10,787 structures. Less than 10 % of PDB ligands (1,453) are present in more than 5 structures and less than 2 % (258) are in more than 55 structures, demonstrating a manageable number of molecules that frequently appear in protein structures (Fig. 1 ). A valuable list composed by Vedadi and col-leagues contains 143 common physiologically relevant molecules, many of which overlap with the most highly represented PDB ligands [ 8 ]. This library provides a good option if the objective of the screen is to identify unknown biological ligands or to verify

2.2 Buffer System

2.3 Screening Libraries

Fluorescence Thermal Shift Screening

268

that a recombinant protein has the capacity to bind its annotated cofactors, substrates, or inhibitors.

In some cases, a library composed of common biological molecules is inconsistent with the screening objective. If the exper-iment seeks to address a narrow hypothesis, a target-specifi c library may present a more appropriate screening option. For example, a kinase inhibitor library may provide the most pertinent information when screening a kinase.

A different set of considerations is relevant when the objec-tive of the screen is not to identify ligands but rather conditions conducive to crystallization. In such cases, the screened library should contain common crystallization additives. Alternatively, conditions in a crystallization screen can be directly tested by FTS in advance of crystallization trials. Screens performed for the pur-pose of informing crystallization are typically conducted at the millimolar concentrations relevant for crystallization trials. Results from these screens can serve as a guide for crystallization, allowing the crystallographer to avoid conditions in which the

Fig. 1 Histogram of the number of occurrences of the top 450 PDB ligands which appeared in more than 25 structures from 14,747 ligands in PDB structures. 10,092 out of 14,747 only appear in one structure. The top 21 are the 20 amino acids plus water, which appear in 71,614–86,016 structures. The next population group consists of 28 ligands that are present in1,009–10,787 structures. These include ions and the four nucleotides. Less than 2 % (258) ligands appear in more than 55 structures, demonstrating that a reasonable sized screen-ing library can be constructed from the molecules that frequently appear in protein structures

Chi-Hao Luan et al.

269

protein exhibits unfavorable properties. Conditions that destabi-lize a protein may produce crystallization-interfering heterogene-ity and can be avoided on the basis of FTS screening results. The avoidance of buffer conditions that produce a multiphasic transi-tion, presumably resulting from protein heterogeneity, has also been shown to be an important consideration for crystallization success [ 16 ].

Because of the relatively high cost of developing the original instru-ment by 3-Dimensional Pharmaceuticals [ 1 ], other devices have been repurposed for FTS assays. Real-time PCR instruments are naturally suited for this assay and were tested by different groups [ 17 ]. They have become widely adopted as the instrument of choice for FTS. Commercially available qPCR machines used to perform FTS assays include the iQ5, CFX96, and CFX384 models by Bio-Rad; the Applied Biosystems 7500 by life Technologies; the LightCycler 480 by Roche; and the Stratagene Mx3005p, to name a few. All Real-time PCR instruments are enabled with (1) 96- or 384-well thermal cycler that can raise the temperature of the sam-ples in discrete, preprogramed steps; and (2) a fl uorescence detec-tor that has the capacity to illuminate each sample with a specifi ed wavelength and detect the fl uorescence emitted by the excited fl uorophore. Although some older qPCR machines, such as the iQ5 and the Applied Biosystems 7900, need to be customized to be able to perform FTS with SYPRO Orange, new models have the capability out of the box.

At the Center for Structural Genomics of Infectious Diseases (CSGID), a TTP Labtech Mosquito to transfer protein samples, a Beckman Biomek FX for adding screen buffers, and a Labcyte Echo 550 for transferring compounds are used. The Mosquito is a nanoliter liquid handling robot confi gured with a fi xed eight chan-nel head and a fi ve position deck. It is fast and accurate at dispens-ing sub-microliter amounts of protein and requires a very small dead volume, allowing conservation of sample. Its precision is not affected by stabilizing detergents in the buffer. The Biomek FX is outfi tted with a 96 channel pipetting head and is the workhorse for pipetting buffers and crystallography screens. The Echo is a non-contact acoustic liquid handling system that transfers fl uid down to 2.5 nL. It allows very precise additions of small molecule ligands in DMSO or aqueous buffers. It can go from any well on a source plate to any well on a destination plate, and therefore can be used to create dilution curves on assay plates. The Echo, as confi gured, cannot accommodate high levels of detergents and takes signifi cantly longer to transfer microliter volumes to a 384-well plate, and, there-fore, is not used for dispensing protein.

2.4 Detection Instrumentation

2.5 Liquid Handling Robotics

Fluorescence Thermal Shift Screening

270

3 Assay Procedure

This discussion is focused on the high-throughput screening aspects of the FTS methods as applied to protein crystallization and drug discovery. Therefore, the procedures are described using 384- or 96-well plates in screening experiments and compound libraries in DMSO.

1. Criteria for hit selection. Unfortunately, there is not always a straightforward formula for identifying FTS hits. In many cases, a simple cutoff (often 3 standard deviations above baseline T m ) can be employed. However, false negatives are a concern with any criteria. Ligand binding will typically produce positive Tm shifts, but in some cases binding will stabilize the denatured state and result in no shift or negative shifts. Numerous exam-ples of known ligands producing negative shifts have been docu-mented. However, because ligands that produce negative shifts only rarely bind the native protein, they are typically disregarded in the high-throughput context. On the other hand, for drug discovery, a ligand that causes negative shift is also interesting, as it can show specifi c binding.

2. Assay volume and protein concentration. If conserving protein samples is not an issue, screening 2 μg of protein in a 10 μL assay volume consistently produces a robust signal. If protein is scarce and thus its conservation is critical, serial analysis of both protein concentration and assay volume can be tested to allow identifi cation of the minimum conditions that produce adequate signal. Even with plenty of protein, a lower concentration is preferred because, as in all molecular characterizations, inter-ference between testing molecules is possible.

3. DMSO tolerance. When screening a DMSO solubilized library it is important to account for the effect of DMSO on protein stability. Some proteins tolerate DMSO poorly, necessitating a lower concentration. If only a low DMSO concentration is well tolerated for a given amount of protein, assay volume can be varied to accommodate. If pooling compounds, it is important that the cumulative amount of DMSO is within the DMSO tolerance of the protein.

4. Plates. Commercial PCR plates are generally suitable for FTS. However, it is necessary to test a new brand or a new batch of plates to see if an artifi cial transition occurs in the absence of protein. In the past, the authors have encountered plates that generate an artifi cial transition at ~50 ºC comparable in magnitude to the protein transition exhibited in a typical FTS experiment.

5. Fluorescent compounds. Some compounds fl uoresce in the green range, interfering with the SYPRO Orange fl uorescent

3.1 Experimental Considerations Prior to Beginning the Assay

Chi-Hao Luan et al.

271

signal. In these cases Nile Red dye can be used as an alternative.

6. pH effect and optimal pH for screening. Protein stability has complex pH dependence. A pH screen is advisable prior to library screening. Figure 2 demonstrates a variety of pH and salt effects with data obtained on protein samples from the CSGID.

Fig. 2 T m plots for 9 CSGID proteins that exhibit distinct patterns of pH and NaCl dependence of protein stability. Proteins are referred by their locus tags

Fluorescence Thermal Shift Screening

272

7. Promiscuous binders (false positives). Some compounds bind nonselectively to multiple proteins. These are easily identifi ed after a number of unrelated proteins are screened. In these cases, the compounds can be fl agged in the analysis for further confi rmation or discarded. When a hit is identifi ed, especially from a relatively new library, a “counter” assay should be per-formed to verify the specifi city of the binding. This can be done with proteins unrelated to the protein of interest.

8. Hit analysis (component contribution analysis). It is advisable to evaluate hits by scrutinizing all the contributing compo-nents. In other words, for molecules that are ion- carriers, the effect should always be compared to other molecules contain-ing the same ion. An example of this occurred when a hit com-pound was found in a library screen of a Staphylococcus aureus protein of unknown function. The molecule identifi ed was a Zn 2+ carrying pyrithione which gave a large T m shift. Subsequent tests found that Zn 2+ alone could cause a large shift as well, indi-cating that the protein can compete with pyrithione for chelated Zn 2+ . Further examination of the protein found that it was a Zn 2+ carrier. Ca 2+ and Cl − are other examples of ions that can frequently produce sizable thermal shifts. Therefore, dissecting the molecules in the screen library into its components is a necessary step for accurately identifying hit molecules.

9. Ligand binding and T m shift. The effect of ligand binding on T m is not always straightforward. It is especially important to be aware of this complexity when using FTS to confi rm hit molecules found in functional assays or other binding assays. Ligand binding may cause positive, negative, or negligible T m shifts. For example, the Campylobacter jejuni MurA exhibits a 0° shift in the presence of its substrate phosphoenolpyruvate, a −1.5 °C shift in the presence of its substrate UDP-N-acetylglucosamine, and a 9 °C shift in the presence of both substrates.

Different considerations are relevant for FTS screens applied for drug discovery and crystallographic purposes. These stem from differences in the chemical properties of screened libraries and the desired ligand binding affi nity. For drug discovery, libraries are often composed of limited solubility molecules and the intent of the screen is to identify high-affi nity binders. Consequently, it is practical to screen compounds at low- to mid-micromolar (typically screened at 10 μM) concentrations.

A sample 384-well assay protocol is provided below. Compound libraries are typically stored at 10 mM in microplate format and nanoliter robotic transfer equipment is used to prepare screens.

3.2 Protocol for Drug Discovery

Chi-Hao Luan et al.

273

1. Prepare protein assay stock in HEPES buffer at 0.1 μg/μL concentration:

Total Volume protein = Number of wells × 10 μL

2. Add 5,000× Sypro Orange (Sypro-O) to protein. The volume of Sypro-O is calculated as:

Volume sypro-O = Total Volume protein × 0.001

3. Transfer 10 μL protein–dye mix to each well of the assay plate. 4. Add compounds to assay plate using an Echo acoustic liquid

handler. A multicompound pooled format is used, where up to fi ve compounds (20 nL each) are added to each well. If the library is dissolved in DMSO, it is critically important to test the effect of DMSO on protein thermal unfolding prior to screen-ing. For practical purposes, the DMSO concentration used in the screen should produce less than a 1 °C T m shift (half of the 2 °C cutoff used to report an FTS hit). Therefore, in some cases, pool size will be determined by the DMSO tolerance limit. Increasing the assay volume could be considered to help in this regard, e.g., increasing the assay volume to 12 μL would allow adding one more compound to the pool and this translates into a 20 % increase in throughput. Other factors to consider when deciding upon pool size are the goal of the screen, the com-pound library concentration, and the anticipated hit rate.

5. Mix and centrifuge to remove air bubbles. A Scientifi c Industries Multi Microplate Genie plate shaker is used with setting 800 for 5 min and the plate is spun at ~200 × g for 1 min.

6. Seal with an optically transparent PCR plate seal. 7. Acquire data on qPCR instrument. Starting at 10 °C, increase

by 0.5 °C per minute up to 95 °C, reading the plate at every temperature point. As the screen progresses, a narrower tem-perature range can be used to speed up plate detection. To avoid false negatives due to the narrow temperature range, all compounds with a melting curve that deviates from the con-trol can be selected as potential hits.

Alternatively, a HEPES/Sypro Orange solution can be pre-pared for diluting protein or preparing other screen solutions. This works better when preparing a small number of assay wells. In practice, if protein stock concentration is no less than 10 times that of the assay concentration, it is acceptable to use a dilution factor 1000X, i.e., 1 μL SYPRO Orange to 1 mL HEPES buffer.

In contrast to drug discovery screens, screens that are meant to guide crystallization contain highly soluble compounds that are not expected to interact with high affi nity to the protein. In this case it makes sense to screen libraries at the low- to mid-millimolar concentrations that may be relevant for crystallization.

3.2.1 384-Well format assay protocol

3.3 Protocol for Thermal Stability Profi ling

Fluorescence Thermal Shift Screening

274

1. Prepare protein assay stock in 100 mM HEPES buffer at 1-2 μg/μL concentration:

Total Volume protein = Number of wells × 1 μL

2. Add 5,000× SYPRO Orange to protein assay stock. The vol-ume of Sypro-O is calculated as:

Volume sypro-O = Total Volume protein × 0.01,

3. Transfer 1 μL of the protein–dye mix to the assay plate with Mosquito liquid handler.

4. Add screen conditions to assay plate, 9 μL per well for each condition, with Biomek FX.

5. Mix and centrifuge to remove air bubbles. A Scientifi c Industries Multi Microplate Genie plate shaker is used with setting 800 for 5 min and the plate is spun at ~200 × g for 1 min.

6. Seal with an optically transparent PCR plate seal. 7. Acquire data on qPCR instrument. Starting at 10 °C, or 30 °C

below the T m , increase by 0.5 °C per minute up to 95 °C, reading the plate at every temperature point.

1. Prepare protein assay stock in HEPES buffer at 1 μg/μL concentration:

Total Volume protein = Number of wells × 1 μL

2. Add Sypro-O to protein. The volume of Sypro-O is calculated as

Volume sypro-O = Total Volume protein × 0.01,

where the Sypro-O stock concentration is 5,000× and assay concentration is 5×.

3. Transfer 2.5 μL protein–dye mix to assay plate. 4. Add screen conditions to assay plate, 22.5 μL per well for each

condition. 5. Mix and centrifuge to remove air bubbles. A Scientifi c

Industries Multi Microplate Genie plate shaker is used with setting 800 for 5 min and the plate is spun at ~200 × g for 1 min.

6. Seal with an optically transparent PCR plate seal. 7. Acquire data on qPCR instrument. Starting at 10 °C, or 30 °C

below the T m , increase by 0.5 °C per minute up to 95 °C, reading the plate at every temperature point.

3.3.1 384-Well format assay protocol

3.3.2 96-Well format assay protocol

Chi-Hao Luan et al.

275

4 Data Analysis

The drug screening process typically starts with a primary screen, followed by a deconvolution screen on the hit wells (if a pooled compound format is used), then a dose–response series on the hit compounds, and fi nally characterization of confi rmed hits. In car-rying out this process there is no need to collect information by other methods to determine which molecules to pick for the next step; the only factor requiring consideration is the result from FTS experiments, namely T m , the shape of the melting curve, and the binding specifi city. To determine these basic parameters, a simple curve fi tting procedure using either the Boltzmann or double Boltzmann equation is suffi cient. The melting curve of a selected hit molecule can, of course, be reanalyzed by a more detailed analysis and used in corroboration with results from other methods. The scope of this chapter only covers FTS, and therefore no discussion of corroborating FTS data with alternate methods is included.

FTS data is recorded as fl uorescence intensity vs. temperature. T m can be determined from the fi rst derivative of the transition profi le without curve fi tting. The fi rst derivative can be automati-cally calculated using the software package that accompanies most qPCR machines. For simple transitions, using the fi rst derivative is an easy and straightforward method to obtain T m . However, the data quality of the derivative is poor compared to using curve fi tted relative fl uorescence units (RFU). It also does not provide an esti-mate of the relative magnitude of the transition if the transition is multiphasic. The transition width is better defi ned from the curve fi tted RFU. The pre-transition (pre-melting) fl uorescence reading is also very informative. Its value depends upon the folding state of the protein. Exposure of hydrophobic patches in unoccupied bind-ing sites or interfaces can result in high background fl uorescence. Reduced background upon the addition of a small molecule indi-cates a possible specifi c interaction between the small molecule and the protein. Therefore, even in library screening mode, the RFU data is used as the primary data for analysis.

Besides transition temperature, T m , the following parameters can be derived from the thermal scanning profi le: transition width k 1 , background reading F A which indicates the protein’s pre- transition folding state, magnitude of transition Δ F , as well as the characteristics of the transition, i.e., whether it is monophasic or multiphasic.

As discussed previously, the transition profi le can be analyzed either with the RFU reading or the fi rst derivative. The fi rst derivative is a direct way to obtain T m and multi-phasic transitions are more visually apparent. However, F A and Δ F are lost in this analysis.

4.1 FTS Parameters

4.2 Curve Fitting Model

Fluorescence Thermal Shift Screening

276

A ligand that stabilizes the native conformation or promotes the folded conformation will often reduce F A . Under most conditions, particularly when protein concentration has been reduced to con-serve the sample, curve fi tted RFU is used for data analysis.

The transition profi le can be fi tted to two different curve fi tting models. One is the Boltzmann sigmoidal model, F x = F A + ( F B − F A )/(1 + exp(( T m − x)/ k 1 )). Simple transitions can be fi t to this model. The model has four parameters: F A , F B , T m , and k 1 , which have been defi ned above and Δ F = F B − F A . Another parameter is also useful, T m10 , which gives more information about the change in cooperativity of the folding process. T m10 can be calculated as T m10 = T m − k 1 × ln9. Thus it converts the transition width k 1 to a temperature so that Δ T m10 can be directly compared with Δ T m to discern the change in transition characteristics with a ligand, as demonstrated in Fig. 3 . For a biphasic transition, a double Boltzmann model can be used.

Broadening of the transition at low compound concentration is an indicator of a high-affi nity binder. This is especially true when the ligand-bound state has a distinct transition, as shown in Fig. 4 with a penicillin-binding protein. When a mixture of bound and unbound protein exists, the two distinct states both contribute to

4.3 Biphasic Transitions

Fig. 3 Normalized melting curves for a CSGID protein with NaCl and MnCl 2 , dem-onstrating the use of T m and T m10 . The steeper transition with MnCl 2 is shown by a greater Δ T m10 value than that of Δ T m . T m10 converts the transition width k 1 in the Boltzmann model to a transition temperature so that Δ T m10 can be directly compared with Δ T m to discern the change in transition characteristics when a ligand bound

Chi-Hao Luan et al.

277

the FTS profi le, appearing convoluted in the fl uorescence reading and as two peaks in the fi rst derivative plot. Therefore, when a hit is found in a primary screen, a dose-curve exploration needs to put more emphasis on studying the melting curve behavior at low compound concentrations. High-affi nity ligands at concentrations closer to the protein concentration will cause this biphasic transition where a broadening of the transition is clearly visible. This may or

Fig. 4 Biphasic transition indicating high-affi nity binding. The data shown is for a secreted penicillin-binding protein from Corynebacterium diphtheriae NCTC 13129 with penicillin. The protein concentration is 1.8 μM. At close to equal molar ratio, the melting curve exhibits a biphasic behavior, indicative of a high- affi nity binding and a two state transition

Fluorescence Thermal Shift Screening

278

may not associate with a large T m shift. The transition width parameter k 1 from Boltzmann fi tting can serve as an indicator for specifi c binding. The peak value in a plot of k 1 vs. ligand concentration indicates the concentration of ligand at which 50 % of the protein is in the free state and 50 % in ligand bound state.

A discussion of the methods by which FTS can be used to deter-mine K D and other binding properties is beyond the scope of this chapter. A sophisticated analysis and a well-described theoretical framework have been reviewed elsewhere [ 5 ]. In general, caution is advised when interpreting affi nity solely on the basis of FTS data. When enthalpy changes of binding are comparable, rank-ordering compounds on the basis of T m should provide an accurate picture of relative affi nity. However, if there are signifi cant differences in enthalpy changes of binding a rank-ordered list may be misleading.

5 A Case Study: 100 CSGID Proteins Screened Against the ADDit™ Library

The data below are from 100 CSGID proteins screened with the ADDit additive screen library by FTS. The data are examined in terms of the property of the screening conditions and the impact on protein stability and solubility behavior. This can have practical implications for assisting scientists in decisions relating to sample preparation and crystallization practice.

The ADDit conditions are marketed by Emerald Biosystems as a screen kit. The conditions are classifi ed into fi ve categories: (1) Salts/Ions, 24 conditions, (2) Volatile Organics/Solvents, 12 con-ditions, (3) Nonvolatile Organics/Polymers, 24 conditions, (4) Detergents, 12 conditions, and (5) Others (Chelators, Reducing Agents, Linkers, Chaotropes, etc.), 24 conditions. The ADDit conditions have been studied systematically on selected proteins via crystallization trials by McPherson and colleagues. Their effects on proteins as manifested in crystallization experiments were reported [ 18 ]. The thermal shift data examines the effects of these same conditions on a set of proteins using a different approach. Because the CSGID proteins tested were not selected by a set of rules other than that they are targets of interest for CSGID, the results provide general insight into physicochemical properties of proteins under the test conditions.

The FTS data for each condition is presented as a histogram of Δ T m for the 100 proteins. In Figs. 5 , 6 , 7 , 8 , 9 , 10 , 11 , and 12 , the proteins are binned into six categories: (1) “No T m ” indicates pro-teins in which no thermal transition was observed under the condi-tion; (2) proteins that exhibited destabilized behavior with Δ T m <−5 °C; (3) proteins that exhibited destabilized behavior with −2 °C > Δ T m > −5 °C; (4) proteins that exhibited little or no change with −2 °C < Δ T m < 2 °C; (5) proteins that were stabilized with

4.4 Hit Follow-Up

Chi-Hao Luan et al.

279

ammonium sulfate60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4 NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4 NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4 NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4 NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4 NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4 NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0

60

40

20

0

60

40

20

0

60

40

20

0

60

40

20

0

60

40

20

0

60

40

20

0

60

40

20

0

60

40

20

0

60

40

20

0

60

40

20

0

barium iodide

barium chloride

calcium chloride dihydrate

cadmium chloride dihydrate

cesium chloride

cupric chloride dihydrate

holmium chloride

lithium chloride

cobalt chloride dihydrate

gadolinium bromide

lanthanum acetate

Fig. 5 Histogram of Δ T m of 100 CSGID proteins showing effect of 12 salts and ions on protein stability. Nonspecifi c solvent effects account for the majority of the events and follow the Hofmeister series

Fluorescence Thermal Shift Screening

280

lithium sulfate

manganese (ll) chloridetetrahydrate

magnesium chloridehexahydrate

potassium citrate

potassium chloride

samarium bromide,hexahydrate

sodium chloride

sodium malonate

yttrium nitrate

samarium chloride

sodium fluoride

yttrium chloride hexahydrate

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

Fig. 6 Histogram of Δ T m of 100 CSGID proteins shows the effect of 12 salts and ions on protein stability

Chi-Hao Luan et al.

281

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

1-propanol 2-Butanol

Acetonitrile

DMSO

Ethyl Acetate

N,N-Dimethylformamide

trifluoroethanol

2-Propanol

dioxane

Ethanol

Methanol

trifluoro acetic acid

Fig. 7 Histogram of Δ T m of 100 CSGID proteins shows the effect of 12 volatile organics/solvents on protein stability

Fluorescence Thermal Shift Screening

282

1,2,3-heptanetriol 1,3 butanediol

1,4 butanediol

1,6 hexanediol

2,5 Hexanediol

dextran sulfate sodium salt

gamma butyrolactone

1,3 propanediol

1,5-diaminopentane di-HCI

1,8-diaminooctane

6-aminocaproic acid

ethylene glycol

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

Fig. 8 Histogram of Δ T m of 100 CSGID proteins shows the effect of 12 nonvolatile organics/polymers on protein stability

Chi-Hao Luan et al.

283

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

glycerol anhydrous

glycyl-glycyl-glycine

MPD

polyethylene glycol 2000 MME

polyethylene glycol 4000

polypropylene glycol P400 polyvinylpyrrolidone K15

polyethylene imine

polyethylene glycol 400

polyethylene glycol 10000

jeffamine M-600

glycine

Fig. 9 Histogram of Δ T m of 100 CSGID proteins shows the effect of another 12 nonvolatile organics/polymers on protein stability

Fluorescence Thermal Shift Screening

284

NAD ATP disodium salt

N-Acetyl-L-Cysteine

EDTA

spermine tetra-HCl

trimethylamine HCI

urea

DTT

BME

spermine

hexaminecobalt trichloride

guanidine HCI

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

Fig. 10 Histogram of Δ T m of 100 CSGID proteins shows the effect of 12 chelators, reducing agents, and chaotropes on protein stability

Chi-Hao Luan et al.

285

Phenol benzamidine HCI

EPPS

D(+)-sucrose

imidazole

non-detergent sulfo-betaine201

taurine

PIPES

D(+)-glucose monohydrate

xylitol

arginine

betaine monohydrate

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

Fig. 11 Histogram of Δ T m of 100 CSGID proteins shows the effect of another 12 chelators, reducing agents, and chaotropes on protein stability

Fluorescence Thermal Shift Screening

n-Dodecyl-b--D-maltoside LDAO

CTAB

b-Octylb-D-glucopyranoside

Polysorbate 80 (Tween 80)

Zwittergent 3-10

n-Octyltetraoxyethylene

Triton X-100

Lauryl Sulfobetaine(Zwittergent 3-12)

CHAPS

Brij 56 (Polyoxyethylene 10cetyl ether)

ASB-14 (Amidosulfobetaine-14)

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

60

40

20

0NoTM < −5 > 5−4 −3 −2 −1 0 1 2 3 4

Fig. 12 Histogram of Δ T m of 100 CSGID proteins shows the effect of detergents on protein stability. Despite the fact that the hydrophobic portion of a detergent can interact with SYPRO Orange to give a prohibitively high background reading, an intact transition was observed for most proteins with three detergents, zwittergent 3-10, Lauryl sulfobetaine (zwittergent 3-12), and n -Octyltetraoxyethylene

287

2 °C < Δ T m < 5 °C; and (6) proteins that exhibited signifi cant stabilized behavior with Δ T m > 5 °C.

Two types of effects were observed, nonspecifi c solvent effects and specifi c binding effects, where the former were the majority of the events with Δ T m values <2 °C and the latter were more rare but with Δ T m > 5 °C. The data indicate that the overall solvent and solute effects follow the Hofmeister series. The more generally relevant properties of these molecules are discussed here. Before continuing with the discussion, it should be noted that the effects are not a result of the His6 tag on the protein. The His6 tag on the N-terminus is included for purifi cation purposes and was not removed before screens were carried out. Hofmeister effects have been observed for many proteins that do not have His6 tags, even on the elastin-based polypeptide with the sequence poly(VPGVG) that has only Val, Pro, Gly residues and only the N- and C-termini are charged in the 50 kDa molecule [ 19 ]. Generally the stability effect is ascribed to ion and protein interactions mediated by backbone rather than charged groups [ 20 ].

Several generalizations can be made from the large dataset. Cations such as sodium, potassium, barium, lithium, cesium, and magnesium are neutral or are slightly stabilizing. The anions sul-fate, citrate, and malonate have strong stabilizing effects. PEGs, glycine, tri-Glycine, and EDTA rarely infl uence protein stability. Even the well-known protein denaturant urea is quite neutral, while guanidinium HCl (GuaHCl) is more destabilizing. However, there is a small population of proteins to which GuaHCl has a sta-bilizing effect. Figure 11 shows that the three sugars in the ADDit library, d -glucose, d -sucrose, and xylitol, are either neutral or stabilizers for a majority of the tested proteins. Simple alcohols destabilized the majority of the proteins studied.

The multivalent cations cadmium, cobalt, gadolinium, lantha-num, samarium, yttrium, holmium, and copper have a detrimental effect as shown by a high population of Δ T m < −5 °C and No T m for a number of proteins. For the same protein, this behavior is cor-related with the observation on NiCl 2 in another library. Because these proteins all contain a His6 purifi cation tag, the effect could partially be due to the His6 tag to the protein’s structural stability.

The detergent data are shown in Fig. 12 . The FTS with SYPRO Orange is not readily applicable to conditions containing detergents. This is due to the SYPRO Orange paradigm: fl uorescence genera-tion is due to the dye interacting with exposed hydrophobic surfaces of the protein when unfolded. The hydrophobic portion of a deter-gent can interact with SYPRO Orange to give a prohibitively high background reading. It is interesting to note, however, that an intact transition was observed for most proteins with three detergents, zwittergent 3-10, Lauryl sulfobetaine (zwittergent 3-12), and n -Octyltetraoxyethylene. They are destabilizers but only mild ones.

Fluorescence Thermal Shift Screening

288

In protein studies, the purpose of using detergent is not for enhanc-ing stability, but rather reducing aggregation. Importantly, the destabilizing effect is not signifi cantly negative.

Acknowledgments

The authors would like to acknowledge the Center for Structural Genomics of Infectious Diseases (CSGID) funded by NIAD under Contracts No. HHSN272200700058C and HHSN272201200026C.

References

1. Pantoliano MW et al (2001) High-density minia-turized thermal shift assays as a general strategy for drug discovery. J Biomol Screen 6:429–440. doi: 10.1089/108705701753364922

2. Zucker FH et al (2010) Prediction of protein crystallization outcome using a hybrid method. J Struct Biol 171:64–73. doi: 10.1016/j.jsb.2010.03.016

3. Senisterra GA, Finerty PJ Jr (2009) High throughput methods of assessing protein sta-bility and aggregation. Mol Biosyst 5:217–223. doi: 10.1039/b814377c

4. Niesen FH, Berglund H, Vedadi M (2007) The use of differential scanning fl uorimetry to detect ligand interactions that promote protein stability. Nat Protoc 2:2212–2221. doi: 10.1038/nprot.2007.321

5. Zhang R, Monsma F (2010) Fluorescence- based thermal shift assays. Curr Opin Drug Discov Devel 13:389–402

6. Lavinder JJ, Hari SB, Sullivan BJ, Magliery TJ (2009) High-throughput thermal scanning: a general, rapid dye-binding thermal shift screen for protein engineering. J Am Chem Soc 131:3794–3795. doi: 10.1021/ja8049063

7. Pershad K, Kay BK (2013) Generating thermal stable variants of protein domains through phage display. Methods 60:38–45. doi: 10.1016/j.ymeth.2012.12.009

8. Vedadi M et al (2006) Chemical screening methods to identify ligands that promote pro-tein stability, protein crystallization, and structure determination. Proc Natl Acad Sci U S A 103:15835–15840. doi: 10.1073/pnas.0605224103

9. Nettleship JE, Brown J, Groves MR, Geerlof A (2008) Methods for protein characterization by mass spectrometry, thermal shift (ThermoFluor) assay, and multiangle or static

light scattering. Methods Mol Biol 426:299–318. doi: 10.1007/978-1-60327-058-8_19

10. Reinhard L, Mayerhofer H, Geerlof A, Mueller- Dieckmann J, Weiss MS (2013) Optimization of protein buffer cocktails using Thermofl uor. Acta Crystallogr Sect F: Struct Biol Cryst Commun 69:209–214. doi: 10.1107/S1744309112051858

11. Phillips K, de la Pena AH (2011). The com-bined use of the Thermofl uor assay and ThermoQ analytical software for the determi-nation of protein stability and buffer optimiza-tion as an aid in protein crystallization. Curr Protoc Mol Biol. Chapter 10, Unit10 28. doi: 10.1002/0471142727.mb1028s94

12. Stepanenko OV et al (2008) Hydrophobic inter-actions and ionic networks play an important role in thermal stability and denaturation mechanism of the porcine odorant-binding protein. Proteins 71:35–44. doi: 10.1002/prot.21658

13. Sheth PR et al (2010) Expression, purifi cation, stability optimization and characterization of human Aurora B kinase domain from E. coli . Arch Biochem Biophys 503:191–201. doi: 10.1016/j.abb.2010.08.004

14. Sledz P, Lang S, Stubbs CJ, Abell C (2012) High-throughput interrogation of ligand bind-ing mode using a fl uorescence-based assay. Angew Chem 51:7680–7683. doi: 10.1002/anie.201202660

15. Giuliani SE, Frank AM, Collart FR (2008) Functional assignment of solute-binding pro-teins of ABC transporters using a fl uorescence- based thermal shift assay. Biochemistry 47:13974–13984. doi: 10.1021/bi801648r

16. Geders TW, Gustafson K, Finzel BC (2012) Use of differential scanning fl uorimetry to optimize the purifi cation and crystallization of PLP-dependent enzymes. Acta Crystallogr

Chi-Hao Luan et al.

289

Sect F: Struct Biol Cryst Commun 68:596–600. doi: 10.1107/S1744309112012912

17. Lo MC et al (2004) Evaluation of fl uorescence- based thermal shift assays for hit identifi cation in drug discovery. Anal Biochem 332:153–159. doi: 10.1016/j.ab.2004.04.031

18. McPherson A, Cudney B (2006) Searching for silver bullets: an alternative strategy for crystal-lizing macromolecules. J Struct Biol 156(3):387–406

19. Urry DW (1993) Molecular machines: how motion and other functions of living organisms can result from reversible chemical changes. Angew Chem (German) 105:859–883; Angew Chem Int Ed Engl 32:819–841

20. Sedlák E, Stagg L, Wittung-Stafshede P (2008) Effect of Hofmeister ions on protein thermal stability: roles of ion hydration and peptide groups? Arch Biochem Biophys 479(1):69–73. doi: 10.1016/j.abb.2008.08.013

Fluorescence Thermal Shift Screening