Fast Mek1 Hit Identification with TRIC Technology Correlates Well with Other Biophysical Methods

Abstract

The variety and complexity of drug targets are expanding rapidly. At the same time, there is significant interest in exploring a larger chemical space to identify new candidates. Fragment-based screening (FBS) has emerged as a popular alternative to traditional high-throughput screening campaigns to identify such drug candidates. FBS identifies hit fragments that exhibit weak interactions with the target of interest, thereby enabling the rational design of small-molecule compounds from the identified hit fragments, which serve as building blocks. This strategy reduces the number of molecules to screen while also allowing the exploration of a greater chemical space.

Here we use temperature-related intensity change (TRIC) technology to perform FBS against the target MAPK/ERK kinase-1 (Mek1). TRIC describes the change in fluorescence intensity of a fluorescently labeled molecule upon a change in temperature. This intensity variation is dependent on the physicochemical environment in the vicinity of the dye and strongly affected by binding events. Thus, the detection of binding events is independent of mass, making TRIC an ideal tool for FBS.

Using only 150 pmol of labeled Mek1, the authors screened 193 fragments from a prescreened library in less than 1 h of measurement time, leading to 66 hits. Among those hits, they identified more than 80% of the published top hits found using orthogonal techniques. Furthermore, TRIC allowed the identification of fragments that were of poor solubility but could be mistaken as false-positive hits in other methods.

Introduction

Fragment-based screening (FBS) is a well-established strategy in small-molecule drug development, and its application has been expanding since it was first developed decades ago. It screens molecules of a reduced size that are composed of fewer than 20 heavy elements. These molecular building blocks, generally smaller than 0.5 kDa, are referred to as fragments. Combinations of different fragments are used to refine the lead synthesis and generate larger molecules, ¹ allowing for tighter affinities. There are compelling advantages to the application of FBS in the drug discovery field. The use of fragments allows the exploration of a large chemical space, while at the same time significantly reducing the number of required molecules to screen.

Although FBS is widely used in drug discovery, some challenges remain when it comes to building a steady and reliable workflow. ² Fragments are low-molecular-weight compounds and thus require a detection method that is independent of mass. Since fragments show weak affinities and sometimes low solubility at high concentration, instruments need to be able to tolerate compound precipitation or great care is needed during experimentation. These and other parameters of FBS elicit the need for orthogonal and parallel detection technologies to yield reliable hit confirmation.^2,3 For targets without biochemical activity, researchers look for biophysical methods as alternatives, but even with smaller library sizes, fast methods are needed, often a weak spot of biophysical techniques.

Here, we evaluated a biophysical method referred to as temperature-related intensity change (TRIC), which is derived from microscale thermophoresis (MST). ⁴ TRIC technology relies on a target molecule that is labeled with a single fluorophore for the detection of binding events. Any fluorescent molecule displays a correlation between brightness and temperature, a property that is dependent on its physicochemical environment. ⁵ If the temperature modification is highly controlled, a given labeled molecule will always display the same change in intensity over the course of the heating process. For TRIC, an infrared laser is used to induce a highly controlled temperature change by excitation of water molecules.^4,6,7 The correlation between the brightness of the fluorescent probe and the temperature of the solvent is strongly linked to the solvation of the labeled molecule. Therefore, any modification of the physicochemical environment of the fluorophore, for example, a binding event, will modify the observed fluorescence temperature correlation.^4,5 Hence, TRIC is a mass-independent detection technology that is highly suited for FBS. The high sensitivity of the fluorescence-based assay design works with low target concentrations. Not only does this allow the screening of a wide range of affinities (from pM to mM), but it also requires less protein sample. In comparison to the related MST technology, TRIC promises higher throughput by enabling a readout of measurement of a 384-well plate in 30 min. To adapt MST to a higher throughput, the sample format had to be changed to agree with SLAS standards. Measurements in a multiwell plate, in contrast to measurements in thin glass capillaries, do not allow the generation of a steep temperature gradient. Hence, the Dianthus instrument line relies on TRIC alone, not thermophoresis. The excellent comparability to MST as a method, as also demonstrated in this publication, results from TRIC also being a major component of the signal measured in MST instrumentation. ⁷

To test the capability of this emerging technology, we conducted a single-point screening campaign against the well-established target protein mitogen/extracellular signal-regulated kinase (Mek1) from the MAPK signaling cascade. Mek1 is a regulator of the ERK1/2 pathway that regulates proliferation, differentiation, transcription, and development. Aberrant signaling in this pathway can lead to unregulated cell growth, and targeting this cascade is promising to develop anticancer drugs. ⁸ We used a library of 193 fragments that had been identified in an initial virtual screen and previously tested for binding to Mek1 using various methods. ⁹ Selecting this library−target combination would allow a direct side-by-side comparison of TRIC technology and established biophysical methods such as MST, surface plasmon resonance (SPR), co-crystallization, and thermal shift assay (TSA) using differential scanning fluorimetry.

The comparative analysis of a single-dose screen of 193 fragment molecules establishes TRIC technology as a promising candidate for FBS, with sufficiently high throughput, mass-independent detection of binding events, high sensitivity for a broad affinity range, and the added advantage of robustness against and detection of poorly soluble fragments.

Materials and Methods

Mek1 Production

The recombinant protein Mek1 kinase domain was produced following a previously published method. ⁹ Mek1 was stored at 270 µM in 20 mM HEPES, 300 mM NaCl, 2 mM DTT, pH 7.5.

Mek1 Labeling

Free cysteine residues in Mek1 were labeled as per instructions in the RED-MALEIMIDE 2nd Generation labeling kit (NanoTemper Technologies, Munich, Germany). Ninety microliters of 10 µM Mek1 was mixed with 10 µL of 300 µM RED-MALEIMIDE 2nd Generation dye diluted from a stock solution resuspended in DMSO to 600 µM. This led to a final mix of 9 µM Mek1 and 30 µM dye at 5% DMSO in 25 mM Tris, 150 mM NaCl, pH 7.4. Note that the labeling buffer composition deviated from instructions in the kit manual. Labeling was performed for 30 min at room temperature in the dark. Unreacted dye was then removed using a gravity flow size-exclusion column provided by the manufacturer to separate free, unreacted dye molecules below 5 kDa from labeled protein material. The size-exclusion column was equilibrated with the assay buffer: 25 mM Tris, 150 mM NaCl, pH 7.4, supplemented with 0.1% Pluronic F-127. Labeled protein was aliquoted and stored at −80 °C to minimize the number of freeze/thaw cycles.

Fragment Library

Fragments were provided by Sanofi and are described by Linke et al. ⁹ Chemical structures remain confidential.

Affinity Determination for Selected Binders

The stability and function of labeled Mek1 were confirmed by an interaction test with a known reference molecule, EP1 (Sanofi). Ten microliters of a serial dilution series of EP1 in 25 mM Tris, 150 mM NaCl, pH 7.4, 0.1% Pluronic F-127 was transferred to each of eight wells. Ten microliters of labeled Mek1 was added to each well, resulting in a final concentration range from 625 nM to 4.9 nM of EP1 and a final concentration of 10 nM labeled Mek1.The titration of fragment 139 was done in the same way as described for EP1, with the only difference being that the final concentration of the fragment ranged from 1 mM to 488.3 nM.

The acquisition was performed on a Dianthus NT.23PicoDuo instrument from NanoTemper Technologies using 25% LED excitation power, picomolar detector sensitivity off, and a laser on-time of 5 s. TRIC dose−response plots show Fnorm (as for MST data) displayed semilogarithmically against the titrated ligand concentration. Fnorm is calculated by computing the ratio of the mean fluorescence recorded within the time window of 4−5 s after the infrared laser was turned on, by the mean initial fluorescence measured (before turning on the infrared laser).^4,7 The analysis was carried out using the DI.Screening Analysis software (NanoTemper Technologies) to merge the data for the final output of the K_d value, as well as to estimate the response area expected for the interaction of EP1 and Mek1 in single-dose assays.

Liquid Handling Sample Preparation for Screening

The final assay condition in the plate was a volume of 20 µL per well at 10 nM Mek1. The assay buffer composition was 10 mM Tris, 150 mM NaCl, pH 7.4, supplemented with 0.1% Pluronic F-127 and 1% DMSO to ensure fragment solubility. A STARlet system (Hamilton) was used for liquid handling. First, fragments were diluted from the 100 mM stock solution into assay buffer to 4 mM and 4% DMSO. The fragments were further diluted fourfold into the Mek1 target stock, to a final concentration of 1 mM and 1% DMSO. The first compound dilution into assay buffer occurred in a conventional microwell plate (Nunc), while the second dilution step into Mek1-containing assay buffer occurred in the Dianthus 384-well microplate (NanoTemper Technologies, cat. DI-P001), where the measurements took place. Each sample was prepared in duplicate and positioned side by side. At regular intervals (18 times in duplicate over the entire screen), a negative control of labeled Mek1-target alone was included in duplicate to ensure robot pipetting accuracy and act as a reference sample for nonbinding conditions. A duplicate for EP1 (five times in duplicate over the entire screen), also pipetted at regular intervals, was used as a positive control to certify Mek1 stability by probing the response area as established in initial assay development experiments. We prepared four rows on five different 384-well plates to make best use of the four-channel STARlet system. Once prepared, each plate was shaken for 6 min at 2400 rpm, incubated for 20 min, and centrifuged for 30 s at 1000g before beginning the measurement.

Fragment Screening with TRIC

The prepared 384-well plates were placed in the instrument immediately after centrifugation. Data acquisition was done at a 25 °C set temperature using 40% LED excitation power along with a sequence of 5 s of infrared laser irradiation.

Acquired data were analyzed using the DI.Screening Analysis software, using a Z-score threshold of 3 and plate-wise referencing of ligands to the corresponding negative controls on the same plate. Subsequent analysis and comparison of the data with orthogonal techniques were done manually using data tables in Microsoft Excel.

Results and Discussion

Assay Development and Optimization

To demonstrate the capability of TRIC technology for FBS, we selected Mek1, a target that has already been extensively screened in the literature and for which data sets from various established technologies were readily available. ⁹ Since TRIC requires a fluorescent target molecule, we coupled Mek1 to a far-red fluorescent dye (RED-MALEIMIDE 2nd Generation from NanoTemper Technologies). Far-red fluorophores are ideal probes for target labeling since they rarely show a background signal in fragment- and small-molecule libraries.

To start, we established the robustness and stability of the system using a known and well-characterized compound, EP1. We measured the area response signal of Mek1 against EP1 in various buffer compositions using a Buffer Exploration Kit (NanoTemper Technologies; data not shown). The best buffer was selected based on the response in amplitude ( Fig. 1A ). We furthermore evaluated the reproducibility of K_d measurements for EP1 binding to Mek1 over the course of 2 h to validate the stability of labeled Mek1 for the duration of the planned screening ( Fig. 1C ). This included the time needed for plate preparation in the liquid handling system and additional shaking, incubation, and measurement time. The ideal assay buffer consisted of 25 mM Tris, 150 mM NaCl, pH 7.4, supplemented with 0.1% Pluronic F-127.

Figure 1.

(A) TRIC traces in a single-point screening scenario. The green curves correspond to Mek1 alone with DMSO as a reference, and the yellow to Mek1 in the presence of EP1. The area between the reference and ligand trace, illustrated in blue, is used as a measure for the response signal. (B) Area response values were calculated for 5 areas between the reference (DMSO) TRIC traces (green), 5 areas between the positive control (EP1) TRIC traces (yellow), and 20 areas between the reference and positive control TRIC traces (purple). Data are shown as a box plot with median (horizontal line within the box) and mean (square within the box). The box is determined by the 25th and 75th percentiles. The whiskers are determined by the 5th and 95th percentiles. Data points (x) denote the maximum and minimum values. (C) Dose−response curve obtained from TRIC trace analysis for EP1 titration series (two time points with n = 2) against labeled Mek1. A duplicate of EP1 was measured just after preparation of the samples (dark yellow) and after 2 h of incubation time (light yellow). A global Langmuir fit to all four individual repeats yielded a K_d of 53 nM, as expected from previous studies.

Overall, maleimide labeling of Mek1 in combination with buffer screening quickly resulted in excellent assay quality as determined by titration of EP1, a known binder. The established conditions were used in a single-point screening of 193 fragment molecules with a low sample consumption of just 150 pmol of labeled protein.

Single-Dose Screening and Evaluation

We screened 193 fragments against Mek1 using one single high dose of 1 mM fragment concentration. We included the well-characterized reference compound EP1 at a saturating concentration of 500 nM every 24 fragments (n = 10) and a Mek1 target-alone negative control every 6 fragments (n = 36). The data were analyzed using the same area analysis approach as described above ( Fig. 1A ). Automated analysis was conducted to identify all compounds that showed a response, exceeding a Z score >3. The median response and global standard deviation from all references measured were computed using DI.Screening Analysis software. The fast measurement time of 1 h could have been further reduced to a little over 30 min with a higher-throughput liquid handling solution. The resulting data, especially the consistently measured negative and positive controls tested five times in duplicate throughout the screening, showed good reproducibility and high assay quality ( Fig. 1B ).

The analysis resulted in the identification of 66 hit fragments with a signal response exceeding the set threshold ( Fig. 2A ). We also identified 48 nonbinders for which no notable change in the measured TRIC response was observed. The remaining 79 fragments fell into other categories that are automatically determined by the DI.Screening Analysis software based on best practices for the analysis of TRIC and MST data ( Table 1 ).

Figure 2.

(A) Pie chart representing the automated ligand categorization resulting from DI.Screening Analysis software. (B) TRIC traces acquired for fragments 80 (dark blue, n = 2) and 55 (cyan, n = 2), which were identified as being autofluorescent (increased initial fluorescence observed for fragment 55) and/or causing target aggregation (aberrant TRIC traces observed for fragments 55 and 80). A Mek1 DMSO reference is shown for comparison (green, n = 2). (C) Relative fluorescence distribution along the X axis of the Dianthus instrument well scan. The fluorescence distribution within the corresponding wells for fragments 55 (cyan, n = 2) and 80 (dark blue, n = 2) shows a striking deviation from the well scans obtained for a DMSO reference (green, n = 2). The shoulders of the observed bell-shaped curves indicate precipitated, fluorescent material. (D) Dose−response curves obtained for fragment 139 in a Dianthus instrument using 2nd Generation RED-MALEIMIDE-labeled Mek1 (left, n = 3) and in a Monolith NT. Automated instrument using RED-NHS-labeled Mek1 (right, n = 1). Data shown in the right panel was kindly provided by the authors of the original MST publication. ⁹ Both instruments show two distinct transitions. The data were grouped into a high-affinity concentration range (purple) and a low-affinity concentration range (gray). When fitted with a Langmuir model, both data sets yield highly comparable affinities.

Table 1.

Ligand Categories Automatically Assigned by DI.Screening Analysis in Single-Dose Screens.

Main Category	Subcategory	Determining Factor
Hit		None of the categories below apply and a significant TRIC signal response is observed.
Nonbinder		None of the categories below apply and no significant TRIC signal response is observed.
Aggregation		The TRIC traces obtained for this ligand show an inhomogeneous pattern that is caused by aggregated or insoluble material in the sample.
Autofluorescence		The steady-state fluorescence for this target−ligand sample significantly exceeds the mean of the reference sample, typically indicating ligand autofluorescence or binding. User intervention and an additional control is required to discern binding from artifact.
Inconclusive	Scan anomaly	The well scans show an aberrant shape that is usually caused by precipitated and aggregated material accumulating at the well bottom,
Inconclusive	Potential hit	Either the steady-state fluorescence for this target−ligand sample is significantly lower than the mean of the reference sample, or ligand-induced time-dependent photobleaching is observed in the TRIC trace. These observations typically indicate spectral properties of the ligand, e.g., causing an inner filter effect. On the other hand, the cause could also be a binding event that strongly alters the spectral properties of the fluorophore. User intervention and an additional control is required to discern binding from artifact.
Inconclusive	Not reproducible	Replicates of the tested ligand do not yield the same category. For example, one replicate shows aggregation, while the other replicate does not.
Inconclusive	Inhomogeneous signal	Inconsistent replicate measurements in TRIC response. These are usually caused by liquid handling errors or sample inhomogeneity, caused by, e.g., aggregated material.

In a typical workflow, additional control experiments for the potential hit and autofluorescence categories would yield a final hit list that would then be used for a secondary K_d screening. While a dilution series of fragments requires more data points than a single-dose screening, it allows for a more detailed and thorough separation of nonspecific and specific binders. However, a secondary screen was beyond the scope of this study, which was designed to test for the comparability of hit identification between different biophysical techniques.

Comparison to Other Biophysical Techniques

For detailed data comparison and interpretation, we followed the approach of the previously published study and focused on the top 25 fragments identified by each orthogonal technique used to characterize Mek1. ⁹ We made use of the established ranking based on the resonance unit (RU) signal from SPR, K_d ranking from MST, ΔT_m ranking from TSA, and co-crystallization successes.

From the top 25 fragments identified by either MST, TSA, or SPR, TRIC technology identified more than 80% as interactors, demonstrating an excellent rate of hit identification by TRIC with low rates of false-negative results. Interestingly, some of the SPR hits with a high RU response showed aberrant well scans in TRIC measurements and also TRIC traces, biased by aggregated target material. For example, fragment 80, which featured a high RU response in SPR, along with a ΔT_m of −1 °C in TSA, ⁹ was identified as aggregated and did not yield a reproducible TRIC response (Fig. 2B,C). Furthermore, this fragment was not identified as a hit in the crystallization study. ⁹ Another fragment, 55, which was a hit identified by both MST and TSA, ⁹ showed strong autofluorescence in TRIC experiments ( Fig. 2B ), coupled to scan anomalies and aggregation (Fig. 2B,C) in the TRIC study. This particular compound was not identified by SPR and did not crystallize with Mek1. ⁹ In conclusion, beyond efficient detection of hit fragments, TRIC could efficiently highlight and automatically categorize fragment molecules of low solubility and with target aggregation-inducing properties. Such fragments are not easily discerned from hits by other technologies, and sorting out such fragments early in the workflow can be time- and money-saving in the long run.

Among the 13 fragments that successfully co-crystallized in the original publication, ⁹ TRIC succeeded to identify all, except fragment 48. This fragment was highly ranked in the MST assay and TSA, but not in the SPR assay. We speculate that the corresponding fragment stock solution might have undergone degradation and thus did not yield a hit in the TRIC assay.

In summary, TRIC technology was used in a single-dose screening with duplicate measurements of one fixed high compound concentration. Nevertheless, a secondary K_d analysis would typically follow, also using TRIC, as illustrated during the assay optimization using EP1. A follow-up K_d screen would enable a more thorough distinction of binders and nonbinders but would increase the required number of data points to be screened.

Nevertheless, to further demonstrate the excellent comparability of TRIC measurements in a Dianthus instrument to MST measurements in a Monolith instrument (both NanoTemper Technologies), we performed a titration series experiment for fragment 139, which showed the highest affinity in the published MST screen. ⁹ Comparison of the titration data obtained on Dianthus ( Fig. 2D , left) was in excellent agreement with the titration data from the previously published MST study ( Fig. 2D , right, raw data kindly provided by the authors of the original publication). In both instruments, the fragment binding to Mek1 resulted in two separate transitions in the lower and higher micromolar concentration ranges. The inverse directionality of the obtained Fnorm data can be attributed to the different labeling strategies and fluorophores used for Mek1 labeling in the two studies. Despite different labeling strategies, different fluorophores, and different instrumentation and detection technology being used, the resulting affinities are in agreement, exemplifying the high comparability between MST and TRIC measurements.

The data presented here clearly demonstrate that TRIC is suited for FBS with some key advantages. The technique is mass-independent and highly sensitive for binding detection in a wide affinity range. Furthermore, the method is buffer-independent, suggesting that it can be used even on unpurified targets, if there is an option for labeling specificity, such as His-tagging.^10,11 This may open the scope to more challenging targets, which are not easily purifiable. Finally, TRIC is highly comparable to and in agreement with other biophysical screening methods, especially MST.