Hot-Spotting with Thermal Scanning: A Ligand ... - ACS Publications

May 24, 2017 - Innovative Medicines and Early Development Biotech Unit, Discovery Sciences, AstraZeneca R&D Gothenburg, 43183 Mölndal,. Sweden...
0 downloads 0 Views 1MB Size
Article pubs.acs.org/jmc

Hot-Spotting with Thermal Scanning: A Ligand- and StructureIndependent Assessment of Target Ligandability Molly Chilton,† Ben Clennell,‡ Fredrik Edfeldt, and Stefan Geschwindner* Innovative Medicines and Early Development Biotech Unit, Discovery Sciences, AstraZeneca R&D Gothenburg, 43183 Mölndal, Sweden S Supporting Information *

ABSTRACT: Evaluating the ligandability of a protein target is a key component when defining hit-finding strategies or when prioritize among drug targets. Computational as well as biophysical approaches based on nuclear magnetic resonance (NMR) fragment screening are powerful approaches but suffer from specific constraints that limit their usage. Here, we demonstrate the applicability of high-throughput thermal scanning (HTTS) as a simple and generic biophysical fragment screening method to reproduce assessments from NMR-based screening. By applying this method to a large set of proteins we can furthermore show that the assessment is predictive of the success of high-throughput screening (HTS). The few divergences for targets of low ligandability originate from the sensitivity differences of the orthogonal biophysical methods. We thus applied a new strategy making use of modulations in the solvent structure to improve assay sensitivity. This novel approach enables improved ligandability assessments in accordance with NMR-based assessments and more importantly positions the methodology as a valuable option for biophysical fragment screening.



INTRODUCTION Failures during the drug discovery process are becoming increasingly expensive and are major drivers for the escalating costs of developing new drugs.1 Reasons for failures can be multifaceted, but a significant fraction particularly affects the early discovery phase.2 In these early stages, they often relate to low chemical tractability as reflected by unsuccessful lead generation approaches or low quality leads especially for new drug target classes.3 The early assessment of the associated drug target risks are thus of upmost importance to positively affect attrition rates through rational risk balancing and by adopting target-specific lead generation approaches to increase the chances of generating high quality leads. Assessing the target druggability, i.e., the ability to modulate a given target by a small molecule with drug-like properties and thereby achieve a therapeutic effect, has subsequently developed into a key activity since it greatly increases our understanding of the chemical tractability of prospective targets entering drug discovery portfolios.4 The druggability of a target is a composite parameter that includes important factors relating not only to effective target engagement but also to desirable pharmacokinetic, pharmacodynamic, and toxicological properties, which actually might only become apparent once the project is more advanced. A necessary, but not sufficient, component of druggability is the propensity of the drug target to bind to small-molecule compounds with high affinity. This key component of druggability has been coined “ligandability” and focuses solely © 2017 American Chemical Society

on molecular recognition and properties of the target binding site.5 This somewhat simplified view of druggability serves as an effective measure for predicting lead generation success due to its foundations on chemical tractability. An efficient method that could predict target ligandability early during the target selection process would subsequently enable more informed decisions about the most appropriate hit finding strategy for a given target and to pursue those most likely to succeed with high priority. Whereas structure-enabled computational methods are frequently applied to predict target druggability,6,7 the particular attractiveness of ligandability relates to the possibility of an experimental assessment with modest layout. This tactic helps to address some of the major shortcomings of computational approaches associated with structurally poorly characterized target classes, the influence of protein dynamics on binding pocket topology, as well as the applied docking algorithm. A variety of experimental approaches based on the screening of compound libraries have thus evolved, and with the advent of fragment-based drug discovery (FBDD), the screening of fragment libraries has seen enhanced utility of the assessment of ligandability.8 By using fragment- instead of compound-libraries consisting of drug-sized molecules, a much improved sampling of chemical space can be achieved, which permits a substantial reduction in the number of compounds Received: February 20, 2017 Published: May 24, 2017 4923

DOI: 10.1021/acs.jmedchem.7b00208 J. Med. Chem. 2017, 60, 4923−4931

Journal of Medicinal Chemistry

Article

required for screening.9 As particular fragments can effectively engage with surface hot spots, i.e., regions of the protein that make a disproportionately large binding energy contribution to a molecular interaction, they are poised to serve as ideal probes for ligandability.10,11 In fact, computational analogues have also recently been found to have good applications for druggability assessment.7 Although the lower molecular size and complexity of fragments enable increased coverage of chemical space, it comes at the cost of reduced affinity, which necessitates the use of sensitive detection methods. Robust biophysical methods such as NMR and surface plasmon resonance (SPR) have traditionally been used to enable reliable detection of fragment binding.12−14 Hajduk et al. introduced NMR screening in conjunction with structural data to quantitatively assess protein targets with druggability indices.15 Although NMR screening has very low false positive and negative hit rates, ligandobserved NMR requires well-characterized tool compounds to conduct competition experiments as means to assess the specificity of the observed primary binding event.16 Known binders with NMR-compatible affinity properties are not always available, especially for novel, unprecedented targets, thereby limiting NMR’s potential for ligandability assessments. Very frequently, there is also no access to the required amounts of (stable-isotope labeled) protein reagents, which might exclude novel disease-relevant targets from such an analysis. SPR has much lower requirements on protein amounts but still requires tool compounds to validate that the protein once tethered to the biosensor is still ligand-binding competent.17 The tethering process presents another challenge, as it sometimes requires tailored protein constructs or is not achievable in terms of preserving the binding competency.18 Consequently, a biophysical method that enables fast and effective fragment screening with sufficient sensitivity to correctly assess target ligandability in concert with minimized constraints on protein reagents and access to tool compounds is highly desirable. This has resulted in a steadily growing toolbox of complementary biophysical approaches used for fragment screening, each offering different throughput and sensitivity. Particularly, the thermal shift assay (TSA) has gained great momentum in that context due to its simplicity and ease of use and is nowadays among the most popular techniques used for fragment screening.13 Nevertheless, it is important to note that the technology is generally experiencing a higher propensity for false positive and false negative hits as, for example, more sophisticated methods like NMR and SPR. This work describes the application of thermal scanning, also known as TSA,19 differential scanning fluorimetry (DSF),20 or ThermoFluor,21 as an alternative means to obtain a ligand- and structure-independent assessment of target ligandability through experimental fragment screening. As this is a platebased technology offering high-throughput data acquisition and analysis, we will use the term high-throughput thermal scanning (HTTS).22 It measures the thermal stability of a protein and the accompanying increase in the protein melting temperature upon ligand binding, which is referred to as the thermal shift. As the magnitude of the thermal shift is not only defined by the ligand affinity but also by a range of other factors including the protein unfolding enthalpy, the ligand binding thermodynamics, and the binding to the unfolded protein species, not all binding events will necessarily lead to a detectable thermal shift.23−26 The appearance of a significant thermal shift is thus typically interpreted as constructive target engagement and

serves as an undemanding and clear-cut hit identification approach that goes hand in hand with low reagent consumption. By performing HTTS using a small fragment library against a large set of drug targets with known lead generation success, we could demonstrate the ability of HTTS to predict the ligandability and subsequently the success of HTS. This is a direct effect of the good reproducibility of the ligandability assessments performed with NMR. For those targets where we could observe discrepancies in the assessment that may be attributed to the lower sensitivity of HTTS, we successfully employed a novel strategy that increases the ability of HTTS to identify low-affinity hits. Through a combined strategy that alters simultaneously the activity as well as the structure of the water network, we provide evidence that HTTS is able to identify an increased number of fragment hits. This results in a more accurate ligandability assessment and simultaneously provides alternative screening tactics for augmenting FBDD campaigns with increased choices of fragment starting points.



RESULTS AND DISCUSSION To assess the predictive power of fragment screening with HTTS, we selected 16 in-house drug targets spanning a range of different target classes for the subsequent ligandability assessment. The selection of the drug targets was driven primarily by prior knowledge of HTS outcome, documented progression from hit-to-lead by any means (e.g., fragment screening, DNA-encoded library screening, new modalities, etc.), and available ligandability assessment by NMR in combination with sufficient amounts of high-quality protein reagents. Particularly, the unavailability of protein reagents restricted the recruitment of the majority of targets from the originally published ligandability study including 36 targets;5 thus, the reduced set was complemented with a small number of novel targets meeting the above requirements. Using this target set, we screened a generic fragment library of 763 compounds using HTTS at a concentration of 1 mM and used the observed hit-rates to assign a ligandability score to the respective targets. The composition of this specific fragment library, which is also denoted as the “ligandability set”, is described in more detail by Fuller et al. and follows improved design principles to maintain and increase success in FBDD campaigns.27 While the general fragment library has seen some significant changes in its composition in the last couple of years, the “ligandability set” remained relatively constant with fewer than 15% of the fragments substituted with analogous substances since the publication of the original ligandability study. This helped to ensure comparability with previously obtained ligandability assessments using NMR and the new study using HTTS. Besides using two technologies with different readouts and sensitivities that will eventually lead to variations in false positive and false negative hit rates, it should be also noted that NMR assessments are based on a number of parameters including ligand affinity and structural diversity as well as primary hit rates as described by Edfeldt et al.5 To intentionally simplify the process for ligandability assessments with HTTS, we only considered the primary hit-rates as classification criteria. More specifically, targets experiencing hitrates below 1.5% have been classified as low ligandable, targets with hit-rates between 1.5% and 4.5% are categorized as medium ligandable, and targets exceeding hit-rates of 4.5% are ranked as highly ligandable. The results of this scoring are presented in Table 1, further detailing information about HTS 4924

DOI: 10.1021/acs.jmedchem.7b00208 J. Med. Chem. 2017, 60, 4923−4931

Journal of Medicinal Chemistry

Article

Table 1. List of 16 Targets Used in This Study, with Ligandability Scores and Outcome of HTS and Success of Hit Finding by Any Means as Well as Target-Specific Details from HTTS and DSC Unfolding Experiments

a

NMR Ligandability scores are according to Edfeldt et al.5 bMelting temperatures are mean values of 16 replicates; errors represent the standard deviation from the mean. cProtein unfolding enthalpies have been determined with DSC under conditions identical to those of HTTS.

assessment from HTTS with the success in HTS, it is apparent that all targets that have been assigned with a low ligandability score have failed to deliver a valuable HTS output for further chemistry progression. An overview of this comparison with HTS outcome is shown in Figure 1. Conversely, the vast majority of targets (>70%) that are deemed ligandable (medium-high ligandability score) by HTTS have in fact seen a successful outcome of their HTS with multiple series to work on during lead generation. It should be noted though, that the set contained a much larger proportion (69%) of targets with unsuccessful HTS outcome that is not necessarily representative of overall HTS success-rates across drug target portfolios (typically >50%); thus, this fraction is expected to change when increasing the number of successful HTS-targets in this set. Another noticeable observation is the low population of targets with a medium ligandability score, which is in stark contrast to the assessment with NMR as seen from Table 1 as well as the previously published ligandability study, which both show a well-balanced distribution between the different categories. This could be a direct consequence of using multiple parameters in the NMR assessment instead of relying solely on primary hit-rates for the categorization of targets. Indeed, when including diversity as an additional criterion for the assessment of highly ligandable targets, one target (Ser/Thr Kinase 1) would be likely categorized as medium ligandable. Nonetheless, it is evident, that despite the intended simplicity of the assessment with HTTS based on primary hit-rates, a low

outcome, project success in lead discovery, as well as the assessment by NMR. As elevated temperatures are known to have a pronounced effect on fragment binding and typically result in reduced affinities, we wanted to make sure that our assessment is not biased by significantly deviating melting temperatures of the target proteins. The average melting temperature for targets with low ligandability is 52.2 ± 6.2 °C, while the average melting temperature for targets with a medium-high ligandability score is 48.4 ± 3.3 °C, which indicates both a low variability and closely matching melting temperatures. Thus, any inability to identify true fragment hits can most likely not be attributed to differences in the melting temperature when using this target set. As not only the affinity but also the protein unfolding enthalpy ΔHunf is influential on the magnitude of the thermal shift, thereby affecting directly the identification of fragment hits by HTTS,23 we also performed measurements with DSC to quantitatively assess ΔHunf. Targets with low ligandability displayed a somewhat lower average value for ΔHunf (121.7 ± 75,8 kcal/mol) as compared to targets with medium-high ligandability (ΔHunf = 154.2 ± 30.3 kcal/mol). As lower values of ΔHunf would rather favor an increase in the thermal shift, thereby positively affecting assay sensitivity, we can exclude that scoring targets with low ligandability is biased by differences in the protein unfolding enthalpy. Ligandability Predicted by HTTS and NMR and Drug Discovery Success. When comparing the ligandability 4925

DOI: 10.1021/acs.jmedchem.7b00208 J. Med. Chem. 2017, 60, 4923−4931

Journal of Medicinal Chemistry

Article

to enter hit-to-lead activities, and interestingly, they have all all been assessed with a low ligandability score, which warrants further analysis. A shared element in those projects was the design of a parallel, alternative hit-finding approach by conducting a fragment screening campaign in support of FBDD. In contrast to the regular experience with FBDD, it was not straightforward to identify a good number of diverse fragments that could be evolved to molecules meeting the lead optimization criteria of those projects. Therefore, while FBDD is frequently able to identify tractable hits against targets for which HTS has failed, this was not the case in these 3 projects which can be linked to the results of the ligandability screen. The low hit-rates in the limited fragment screen used for the ligandability assessment were thus not only a pointer toward failure in HTS but also indicated a higher risk of pursuing FBDD as an alternative hit finding option. With this in mind, there is still a need to weigh in other aspects as well, relating to the chemical diversity and attractiveness of the identified hits, their ligand efficiency, and the readiness of the established structural system for iterative support in order to assess the overall risk of a FBDD campaign. If such an assessment turns out to indicate an elevated risk for FBDD, it will be critical to include other hit-finding or diseasemodifying approaches as well, particularly for high-value drug targets. Such approaches may include but are not limited to DNA-encoded library screening,28 which allows for screening of much larger libraries as feasible for HTS or FBDD, or alternative modalities, that go beyond small molecules. The latter include, for example, cyclic or stapled peptides, antisense oligonucleotides, or modified mRNA to name just a few members of the new modalities space.29 In cases where an elevated risk for FBDD is connected to a low hit-rate observed in the experimental ligandability assessment with HTTS, it could be a consequence of the insufficient assay sensitivity required to detect weak fragment binding. Hence, we also compared the ligandability scores obtained by the classical NMR approach with the assessment based on HTTS hit-rates (Figure 3). All of the targets that have been assigned either with a low or a high ligandability score by employing NMR yielded the same assessment as with HTTS,

Figure 1. Results from the experimental ligandability assessment of 16 targets with HTTS, binned according to their ligandability score and color-coded corresponding to HTS success.

ligandability score is very strongly coupled with unsuccessful HTS, whereas targets with increased ligandability display a clear trend for a higher probability of delivering chemical starting points for medicinal chemistry optimization. The findings correspond well with the common observation that improved hit rates in fragment screening often reflect enhanced chemical diversity, which will consequently increase the probability of finding structural analogues in internal or external compound collections via HTS. However, since the entry into hit-to-lead activities is not only dictated by the success of HTS but also influenced by alternative lead generation approaches, the comparison of the ligandability assessment from HTTS and the ability to engage in hit-to-lead activities provides a rather different picture as depicted in Figure 2. Although more than two-thirds of the projects in that target set did not see a successful HTS, this did not translate into a complete stop of lead generation activities in most of those projects. In fact, only 3 projects could not be progressed

Figure 2. Results from the experimental ligandability assessment of 16 targets with HTTS, binned according to their ligandability score and color-coded corresponding to the ability to enter into hit-to-lead activities.

Figure 3. Comparison of the experimental ligandability assessment scores of 16 targets with NMR and HTTS. Targets are binned according to their ligandability score with NMR and color-coded corresponding to their ligandability score with HTTS. 4926

DOI: 10.1021/acs.jmedchem.7b00208 J. Med. Chem. 2017, 60, 4923−4931

Journal of Medicinal Chemistry

Article

Table 2. List of 3 Targets with Low Ligandability Scores under Standard Solvent Conditions and the Impact of the Change in Solvent Conditions on the Observed Hit-Rates by HTTS as Well as the Associated Ligandability Scores

established that change in solvent organization is one of the main factors influencing thermodynamic binding profiles.33,34 Thus, changing the solvent organization should provide an effective handle to manipulate the thermodynamic profile of fragment binding, in particular as water molecules that are displaced by fragments show a differentiating thermodynamic profile from those displaced by drug-like molecules.35 Subtle changes in the solvent organization can be readily achieved through a substitution of H2O with D2O. The changes in the water network originate from quantum differences between light and heavy water that lead to small alterations in the hydrogen-bond length.36 Moreover, such substitutions have been frequently used in the past to study thermodynamic solvent isotope effects and typically lead to a decrease in the enthalpy of binding and consequently an increase in the entropic contribution as the affinity in D2O remains essentially unchanged.37 As we expect this to be even more pronounced for fragments, we evaluated the impact of solvent isotope effects on the ability to identify fragments hits with HTTS and consequently the ligandability assessment for those 3 targets. We rescreened the same generic fragment library against the 3 selected targets in D2O-containing buffer and again used only the observed hit rates to assign a ligandability score following principles similar to those previously outlined. As expected and also documented in Table 2, there is the general trend of much higher hit rates (at least 3-fold increase) in D2O when compared with the original hit rates in H2O. This appears to be a direct consequence of an increased entropic component to the overall fragment binding energy due to the discussed solvent isotope effects but can in part be attributed to an occasional increase in affinity through the formation of stronger deuterium-bonds with the binding site. While this was sufficient to change the ligandability assessment for 2 of the targets, the assessment for the third target remained unchanged. As protein-unfolding in D2O is known to have profound effects on protein stability due to poorer solvation of nonpolar amino acid residues, it cannot be excluded that a D2O-induced increase in the unfolding enthalpy impacts negatively on the ability to detect fragment binding.38 The stability increase in D2O-containing buffer accounted in fact for a remarkable 3.9− 4.5 °C shift in the melting temperatures of the respective protein targets and could eventually cancel out some of the entropy-induced thermal shifts for some fragments. We thus made use of the nonionic chaotropic substance urea to counteract these potential effects and to evaluate if one is able to further increase the capacity of HTTS to detect fragment hits. Urea is known to reduce the unfolding enthalpy and thus the stability of proteins through a combination of changes in solvent structure and dynamic properties as well as a higher degree of solvation of nonpolar amino acid residues.39,40 As urea, despite being a fluid phase modifier, can eventually

which is remarkable and positions HTTS as a valuable alternative. Divergences in the assessment are prominently clustered in the medium ligandability score, which is in part a consequence of the boundary definition used in the classification and reflects the fact that this defines a gray area between low and high ligandabilities. Nevertheless, as an incorrect assignment of a low ligandability score would likely trigger a whole cascade of resource-intensive activities in terms of alternative hit-finding approaches, we focused on those 3 targets that got a low ligandability score with HTTS but were assessed with a higher score when using NMR. The primary aim was to evaluate if the lower assay sensitivity of HTTS is indeed causing a low ligandability score in those cases where the apparent ligandability is actually higher and how this can eventually be avoided by applying novel sensitivity enhancement tactics for HTTS. For this to be effective, it is important to understand the full range of tunable thermodynamic factors that contribute to a detectable thermal shift when using HTTS.30 Increasing Assay Sensitivity of HTTS for Fragment Screening. The magnitude of the thermal shift that is observed in a HTTS experiment is not solely dictated by the affinity of the interacting ligand but is also significantly influenced by a range of other factors that include both the enthalpy of protein unfolding as well as the entropy of ligand binding.23−26 As one would intuitively expect, an increase in the ligand binding affinity typically enhances the thermal shift if the thermodynamic profile and thus the binding mode of the ligand remains conserved. Similar correlations between a detectable binding signal and affinity are actually a common denominator for most other biophysical methods that are used for the interrogation of protein−ligand interactions like SPR and NMR. Additionally, the thermodynamic profile of the ligand binding impacts in such a way that entropy-dominated binding will generally lead to a much larger thermal shift as opposed to enthalpy-dominated binding, if the free energy of binding and thus the affinity remains unchanged. Finally, a low enthalpy of protein unfolding that equates to lower protein stability will typically promote an increase in the thermal shift as opposed to more stable protein systems that display a high enthalpy of protein unfolding. Both the thermodynamics of protein unfolding and ligand binding can be influenced by extrinsic factors, which furnishes the opportunity to increase the sensitivity of HTTS especially with regard to fragment screening. Fragments display a noticeable thermodynamic discrimination as opposed to drug-like molecules in their tendency for enthalpically driven binding.31,32 This, together with their low affinity, represents a thermodynamic rationale for their propensity to frequently produce small thermal shifts in HTTS that can be below the limit of detection. It is well 4927

DOI: 10.1021/acs.jmedchem.7b00208 J. Med. Chem. 2017, 60, 4923−4931

Journal of Medicinal Chemistry

Article

binding enthalpy at elevated temperatures or the differential conformational flexibility for targets with a high unfolding enthalpy presenting a reduced number of available conformers for fragment binding. Hence, there is still the need to assess the individual responsiveness of the protein system to this novel screening strategy by varying the buffer and solvent conditions.

modulate protein−ligand interactions in a very protein- and region-selective manner,41 we made use of target-specific reference compounds to define the minimal urea concentration required to achieve a maximal thermal shift increase when compared with that in normal buffer conditions. As predicted and shown in Table 2, the addition of nondenaturing concentrations of urea (between 0.8 and 2.4 M for the respective targets) had effects on the HTTS hit-rates for all 3 targets. The most pronounced could be observed for BACE, where a relative reduction of the protein stability by >14 °C when compared to that with D2O alone leads to a more than 10-fold increase in the HTTS hit rate and thus strongly affects the ligandability score. A much less pronounced but still significant effect could also be observed for the PPI-target, where the hit-rate was moderately increased. Interestingly, for the third target one can observe an opposing trend, but this had no impact on the ligandability score. The physicochemical properties for the fragment hits obtained through the addition of urea showed nearly analogous profiles campared to those found in D2O alone (see Supporting Information, Figure S2). From the plot of clogP, fragment hits found in urea appear to be slightly more lipophilic, and their mean heavy atom count is marginally increased. While the number of hydrogen bond acceptors is somewhat reduced in urea, there was no observable trend for the number of hydrogen bond donors. These findings suggests that the risk of occasional modulation of protein− ligand interactions by urea can be mainly mitigated through the use of target-specific reference compounds that enable to define optimal urea concentrations that do not interfere with fragment binding. Taken together, changes in the solvent organization and network through the combined usage of D2O and urea leads to significant changes in fragment hit-rates in HTTS, which results in ligandability scores that are more closely aligned with those obtained by NMR. This appears to be a direct consequence of the improved HTTS assay sensitivity caused by a simultaneous increase of the ligand binding entropy and reduction of the protein unfolding enthalpy. Interestingly, applying this screening strategy for BACE enabled us to pull out exclusively amidine-containing fragment scaffolds with the vast majority representing the classical BACE key pharmacophore of a nonplanar cyclic amidine.42 This suggests that this novel approach is not artificially increasing false-positive rates, thereby turning low ligandable targets into high ligandable targets. Interestingly, weighing in diversity as another parameter for ligandability assessment would in fact reduce BACE ligandability to medium ligandability, as also previously seen with another target from the set. The method for increasing HTTS assay sensitivity is not only applicable for ligandability assessments but more generally in fragment screening in order to increase the choice of primary hits in FBDD campaigns. While it is acknowledged that detection of fragment binding with HTTS can lead to a variety of artifacts, false positives and false negatives, it is expected that the discussed changes in solvent architecture will be advantageous in that respect.43 In fact, we have used this concept in a couple of FBDD campaigns and have seen improved assay behavior. Something that we nevertheless noticed is the interesting detail that targets with a high temperature of unfolding (>70 °C) are less responsive to this approach as targets that display moderate melting temperatures and thus a lower thermodynamic stability. There might be several reasons for this including further reduced binding affinity and increased



CONCLUSIONS We show here that the ligandability of target proteins can be experimentally assessed with HTTS and can serve as a predictive tool for chemical tractability. In contrast to other experimental as well as computational methods, this approach is highly effective and independent from access to structural information or tool compounds/ligands. Importantly, the assessment of ligandability is uncoupled from binding site knowledge and therefore goes beyond conventional approaches that involve protein−drug complexes or the use functional ligands to assess competitive binding to known drug binding sites. As a result, this approach provides additional information about the ligandability of allosteric (cryptic) binding sites, i.e., the “allostericability” adding an important layer of information to the decision making process in early drug discovery. Notably, the addition of known tool compounds during HTTS can aid the specific detection of allosteric binders through additive effects on the unfolding temperature.44 It will thus inform not only about the existence of such site(s) but also provide a simultaneous assessment of its chemical tractability and inform lead generation tactics. In order to increase the reliability of ligandability assessments with HTTS, we furthermore developed a new method that improves the sensitivity of HTTS. Reduced unfolding and binding enthalpies induced by changes in solvent structure and activity could significantly enhance the ability of HTTS to detect weak fragment binding events. Including structural diversity of the primary hits during the ligandability assessment can provide an additional parameter to further advance the predictability of this approach. While the HTTS sensitivity enhancement seems to enable more robust ligandability assessments in accordance with NMR, it also has general consequences for FBDD by positioning HTTS as a viable biophysical option for larger fragment screening campaigns. The particular attractiveness of this approach results from the moderate technical requirements, the high-throughput and reduced reagent constraints that can even include membrane protein targets for future studies.45 Thus, it can effectively serve as a target-agnostic first-pass fragment screening approach with more sophisticated biophysical methods, such as NMR and SPR, being used for subsequent hit validation and detailed characterization of their mode of action.



EXPERIMENTAL SECTION

Proteins and Reagents. All proteins were heterologously expressed and purified at AstraZeneca using standard procedures. Protein purities were assessed with gel-electrophoresis, and the ligandbinding competence of the individual targets was typically determined via an active site titration with a potent small-molecule inhibitor using HTTS. The ligandability set was synthesized at AstraZeneca; further information about its composition can be found at Fuller et al.27 Buffer components were purchased from Teknova or Sigma-Aldrich unless otherwise stated. SYPRO Orange was purchased from Sigma-Aldrich as a 5000-times concentrated solution in 100% DMSO. D2O (99.9% deuterium incorporation level) was obtained from Cambridge Stable Isotopes. 4928

DOI: 10.1021/acs.jmedchem.7b00208 J. Med. Chem. 2017, 60, 4923−4931

Journal of Medicinal Chemistry

Article

High-throughput Thermal Scanning Assay (HTTS). The HTTS assay was conducted using a LightCycler480 (Roche). Protein unfolding was monitored with the help of the fluorescent dye SYPRO Orange.20 The fluorescence intensity was measured at an emission wavelength of 510 nm and an excitation wavelength of 465 nm. The concentrations of SYPRO Orange and the target proteins were individually optimized. The optimization assay consisted typically of a small matrix made of a gradient with increasing protein concentration against a gradient of increasing SYPRO Orange concentration. Concentrations typically ranged from 0.02 to 0.1 mg/mL for protein and a 1:500−1:1500 dilution of SYPRO Orange. This assay window was found to be sufficient for capturing optimal melting curves for all protein targets. The SYPRO Orange was dispensed using a HP D300 Digital Dispenser (Tecan). The protein and buffers were dispensed using a Mantis liquid handler (Formulatrix). When selecting the optimized conditions for protein consumption, initial fluorescence and signal levels were taken into account as well as the influence of SYPRO Orange on the observed melting temperature. The optimized protein and SYPRO Orange concentrations along with the screening buffers for all selected targets can be found in the Supporting Information. Fragments at 100 mM stock concentration, dissolved in DMSO, were added to Framestar 384 well PCR plates (4titude Ltd.) using an Echo Liquid Handler (Labcyte Inc.) at a volume of 100 nL. The premixed optimized protein and SYPRO Orange mixture was added to the assay ready plates to a total volume of 10 μL yielding a final screening concentration of 1 mM in 1% (v/v) DMSO. A Mantis liquid handler (Formulatrix) or a FluidX Liquid Dispenser was used for dispensing. For each target, the assay was validated by showing a thermal stabilization of the protein in the presence of 100 μM of a known small-molecule binder. The reference compound and DMSO control were added to the assay ready plates using a HP D300 Digital Dispenser (Tecan). The sealed plate was heated from 20 to 85 °C at a heating rate of 2 °C/min using a protein-melting template in the LightCycler480 software. HTTS Assay Sensitivity Optimization. To study the effects of D2O and urea on HTTS assay sensitivity, buffers for the selected targets were prepared from dry buffer substances using D2O (99.9%). The pD was measured and adjusted by using a normal pH-meter and applying a formula for correlating pKa values determined in D2O and H2O.46 For defining the most favorable urea concentration, the optimization assay consisted typically of a small matrix made of a gradient with increasing urea concentration against a gradient of increasing reference compound concentration. The urea concentration that induced the largest thermal shift for the respective reference compound was typically selected for the screening of the ligandability set with HTTS. An important aspect in this selection was also the assessment of the slope of the unfolding curve, as this should be unaffected by the presence of urea. The optimized screening buffer composition can be found in the Supporting Information. HTTS Data Analysis. Data were analyzed using a thermal shift assay template in Genedata Screener (version 12.0.4). Two calculation methods were used to determine the melting temperature (Tm). The geometrical method makes use of the minimum and maximum fluorescence signal before and after the transition to calculate the midpoint of the thermal transition, which is defined by 50% of the protein being transitioned to the unfolded state. The first derivative method uses the maximum of the first derivative as the melting temperature. Melting temperatures from both methods should essentially agree with one another when following a two-state unfolding model, and differences can be used to identify irregular curves for visual assessment and annotation. To report the Tm- and ΔTm-values, those calculated using the first derivative method were used. The ΔTm or thermal shift for each fragment is defined as the difference in Tm from the average of the 16 DMSO controls containing no compound. Hit selection criteria are based on a significant thermal stabilization that should be at least 3-times above the standard deviation of the DMSO controls. The primary screening data for all the targets as visualized with Spotfire software can be found in the Supporting Information.

Differential Scanning Calorimetry (DSC). DSC was performed using a Microcal VP-DSC (Malvern). All protein samples were buffer exchanged with the TSA buffer using Illustra NAP-5 columns (GE Healthcare). The sample cell was loaded with approximately 1 mg/mL of protein and the reference cell with buffer. Samples were heated from 20 to 85 °C at a scan rate of 90 °C/h. Data were analyzed using Origin DSC software (version 7.0). The data were normalized for the protein concentration determined using a Nanodrop 2000 (Thermo Scientific). After a baseline was fitted and subtracted, the curve was aligned using the non-2-state model. This yielded the unfolding enthalpy (ΔHunf) and the melting temperature (Tm) of the protein. For the DSC experiments, the buffer and start and end temperatures were the same as those in the HTTS experiments.



ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.jmedchem.7b00208. Optimized protein and SYPRO Orange screening concentrations used in the HTTS ligandability screen and buffer conditions for HTTS and DSC experiments, buffer conditions and unfolding temperatures for the 3 selected targets for HTTS sensitivity enhancement, primary screening data for all targets, and property comparison of screening hits obtained through HTTS sensitivity enhancement (PDF)



AUTHOR INFORMATION

Corresponding Author

*Phone: +46 31 776 21 97. E-mail: stefan.geschwindner@ astrazeneca.com. ORCID

Stefan Geschwindner: 0000-0002-2154-8345 Present Addresses

† M.C.: School of Biochemistry, Biomedical Sciences Building, University of Bristol, Bristol BS8 1TD, U.K. ‡ B.C.: School of Science & Technology, Clifton Campus, Nottingham Trent University, Nottingham NG11 8NS, U.K.

Notes

The authors declare no competing financial interest.



ACKNOWLEDGMENTS We thank Loredana Spadola for further computational analysis of the fragment screening hits. Furthermore, we thank AstraZeneca for providing support to Molly Chilton and Ben Clennell as industrial placement students within the Structure & Biophysics department.



ABBREVIATIONS USED BACE, β-amyloid converting enzyme; DSC, differential scanning calorimetry; DSF, differential scanning fluorimetry; FBDD, fragment-based drug discovery; HTS, high throughput screening; HTTS, high throughput thermal scanning; NMR, nuclear magnetic resonance; PPI, protein−protein interaction; SPR, surface plasmon resonance; TSA, thermal shift assay



REFERENCES

(1) Pammolli, F.; Magazzini, L.; Riccaboni, M. The productivity crisis in pharmaceutical R&D. Nat. Rev. Drug Discovery 2011, 10 (6), 428− 438. (2) Waring, M. J.; Arrowsmith, J.; Leach, A. R.; Leeson, P. D.; Mandrell, S.; Owen, R. M.; Pairaudeau, G.; Pennie, W. D.; Pickett, S. 4929

DOI: 10.1021/acs.jmedchem.7b00208 J. Med. Chem. 2017, 60, 4923−4931

Journal of Medicinal Chemistry

Article

shift screen for protein engineering. J. Am. Chem. Soc. 2009, 131 (11), 3794−3795. (23) Matulis, D.; Kranz, J. K.; Salemme, F. R.; Todd, M. J. Thermodynamic stability of carbonic anhydrase: measurements of binding affinity and stoichiometry using ThermoFluor. Biochemistry 2005, 44 (13), 5258−5266. (24) Holdgate, G. A.; Ward, W. H. Measurements of binding thermodynamics in drug discovery. Drug Discovery Today 2005, 10 (22), 1543−1550. (25) Holdgate, G. A. Thermodynamics of binding interactions in the rational drug design process. Expert Opin. Drug Discovery 2007, 2 (8), 1103−1114. (26) Cimmperman, P.; Baranauskiene, L.; Jachimoviciute, S.; Jachno, J.; Torresan, J.; Michailoviene, V.; Matuliene, J.; Sereikaite, J.; Bumelis, V.; Matulis, D. A quantitative model of thermal stabilization and destabilization of proteins by ligands. Biophys. J. 2008, 95 (7), 3222− 3231. (27) Fuller, N.; Spadola, L.; Cowen, S.; Patel, J.; Schonherr, H.; Cao, Q.; McKenzie, A.; Edfeldt, F.; Rabow, A.; Goodnow, R. An improved model for fragment-based lead generation at AstraZeneca. Drug Discovery Today 2016, 21 (8), 1272−1283. (28) Goodnow, R. A., Jr.; Dumelin, C. E.; Keefe, A. D. DNA-encoded chemistry: enabling the deeper sampling of chemical space. Nat. Rev. Drug Discovery 2017, 16 (2), 131−147. (29) Waldmann, H.; Valeur, E.; Gueret, S. M.; Adihou, H.; Gopalakrishnan, R.; Lemurell, M.; Grossmann, T. N.; Plowright, A. T., New modalities for challenging targets in drug discovery. Angew. Chem. [Online early access]. DOI: 10.1002/ange.201611914. Published Online: Feb 10, 2017. (30) Layton, C. J.; Hellinga, H. W. Thermodynamic analysis of ligand-induced changes in protein thermal unfolding applied to highthroughput determination of ligand affinities with extrinsic fluorescent dyes. Biochemistry 2010, 49 (51), 10831−10841. (31) Ferenczy, G. G.; Keseru, G. M. On the enthalpic preference of fragment binding. MedChemComm 2016, 7 (2), 332−337. (32) Williams, G.; Ferenczy, G. G.; Ulander, J.; Keseru, G. M. Binding thermodynamics discriminates fragments from druglike compounds: a thermodynamic description of fragment-based drug discovery. Drug Discovery Today 2017, 22 (4), 681−689. (33) Chervenak, M. C.; Toone, E. J. A Direct measure of the contribution of solvent reorganization to the enthalpy of ligandbinding. J. Am. Chem. Soc. 1994, 116 (23), 10533−10539. (34) Geschwindner, S.; Ulander, J.; Johansson, P. Ligand binding thermodynamics in drug discovery: still a hot tip? J. Med. Chem. 2015, 58 (16), 6321−6335. (35) Ichihara, O.; Shimada, Y.; Yoshidome, D. The importance of hydration thermodynamics in fragment-to-lead optimization. ChemMedChem 2014, 9 (12), 2708−2717. (36) Soper, A. K.; Benmore, C. J. Quantum differences between heavy and light water. Phys. Rev. Lett. 2008, 101 (6), 655021−655024. (37) Duff, M. R., Jr.; Howell, E. E. Thermodynamics and solvent linkage of macromolecule-ligand interactions. Methods 2015, 76, 51− 60. (38) Efimova, Y. M.; Haemers, S.; Wierczinski, B.; Norde, W.; van Well, A. A. Stability of globular proteins in H2O and D2O. Biopolymers 2007, 85 (3), 264−273. (39) Bennion, B. J.; Daggett, V. The molecular basis for the chemical denaturation of proteins by urea. Proc. Natl. Acad. Sci. U. S. A. 2003, 100 (9), 5142−5147. (40) Das, A.; Mukhopadhyay, C. Urea-mediated protein denaturation: a consensus view. J. Phys. Chem. B 2009, 113 (38), 12816− 12824. (41) Holstein, M. A.; Parimal, S.; McCallum, S. A.; Cramer, S. M. Effects of urea on selectivity and protein-ligand interactions in multimodal cation exchange chromatography. Langmuir 2013, 29 (1), 158−167. (42) Stamford, A.; Strickland, C. Inhibitors of BACE for treating alzheimer’s disease: a fragment-based drug discovery story. Curr. Opin. Chem. Biol. 2013, 17 (3), 320−328.

D.; Wang, J.; Wallace, O.; Weir, A. An analysis of the attrition of drug candidates from four major pharmaceutical companies. Nat. Rev. Drug Discovery 2015, 14 (7), 475−486. (3) Barker, A.; Kettle, J. G.; Nowak, T.; Pease, J. E. Expanding medicinal chemistry space. Drug Discovery Today 2013, 18 (5−6), 298−304. (4) Bauer, U.; Breeze, A. L.“Ligandability” of Drug Targets: Assessment of Chemical Tractability via Experimental and in Silico Approaches. In Lead Generation: Methods and Strategies; Wiley VCH: Weinheim, Germany, 2016; Vol. 68a, pp 37−63. DOI: 10.1002/ 9783527677047.ch03. (5) Edfeldt, F. N.; Folmer, R. H.; Breeze, A. L. Fragment screening to predict druggability (ligandability) and lead discovery success. Drug Discovery Today 2011, 16 (7−8), 284−287. (6) Egner, U.; Hillig, R. C. A structural biology view of target drugability. Expert Opin. Drug Discovery 2008, 3 (4), 391−401. (7) Kozakov, D.; Hall, D. R.; Napoleon, R. L.; Yueh, C.; Whitty, A.; Vajda, S. New frontiers in druggability. J. Med. Chem. 2015, 58 (23), 9063−9088. (8) Erlanson, D. A.; Fesik, S. W.; Hubbard, R. E.; Jahnke, W.; Jhoti, H. Twenty years on: the impact of fragments on drug discovery. Nat. Rev. Drug Discovery 2016, 15 (9), 605−619. (9) Erlanson, D. A.; McDowell, R. S.; O’Brien, T. Fragment-based drug discovery. J. Med. Chem. 2004, 47 (14), 3463−3482. (10) Ciulli, A.; Williams, G.; Smith, A. G.; Blundell, T. L.; Abell, C. Probing hot spots at protein-ligand binding sites: a fragment-based approach using biophysical methods. J. Med. Chem. 2006, 49 (16), 4992−5000. (11) Radoux, C. J.; Olsson, T. S.; Pitt, W. R.; Groom, C. R.; Blundell, T. L. Identifying interactions that determine fragment binding at protein hotspots. J. Med. Chem. 2016, 59 (9), 4314−4325. (12) Dalvit, C. NMR methods in fragment screening: theory and a comparison with other biophysical techniques. Drug Discovery Today 2009, 14 (21−22), 1051−1057. (13) Keseru, G. M.; Erlanson, D. A.; Ferenczy, G. G.; Hann, M. M.; Murray, C. W.; Pickett, S. D. Design principles for fragment libraries: maximizing the value of learnings from pharma fragment-based drug discovery (FBDD) programs for use in academia. J. Med. Chem. 2016, 59 (18), 8189−8206. (14) Danielson, U. H. Fragment library screening and lead characterization using SPR biosensors. Curr. Top. Med. Chem. 2009, 9 (18), 1725−1735. (15) Hajduk, P. J.; Huth, J. R.; Fesik, S. W. Druggability indices for protein targets derived from NMR-based screening data. J. Med. Chem. 2005, 48 (7), 2518−2525. (16) Holdgate, G. A.; Anderson, M.; Edfeldt, F.; Geschwindner, S. Affinity-based, biophysical methods to detect and analyze ligand binding to recombinant proteins: matching high information content with high throughput. J. Struct. Biol. 2010, 172 (1), 142−157. (17) Kaminski, T.; Gunnarsson, A.; Geschwindner, S. Harnessing the versatility of optical biosensors for target-based small-molecule drug discovery. ACS Sensors 2017, 2 (1), 10−15. (18) Holdgate, G.; Geschwindner, S.; Breeze, A.; Davies, G.; Colclough, N.; Temesi, D.; Ward, L. Biophysical methods in drug discovery from small molecule to pharmaceutical. Methods Mol. Biol. 2013, 1008, 327−355. (19) Pantoliano, M. W.; Petrella, E. C.; Kwasnoski, J. D.; Lobanov, V. S.; Myslik, J.; Graf, E.; Carver, T.; Asel, E.; Springer, B. A.; Lane, P.; Salemme, F. R. High-density miniaturized thermal shift assays as a general strategy for drug discovery. J. Biomol. Screening 2001, 6 (6), 429−440. (20) Niesen, F. H.; Berglund, H.; Vedadi, M. The use of differential scanning fluorimetry to detect ligand interactions that promote protein stability. Nat. Protoc. 2007, 2 (9), 2212−2221. (21) Cummings, M. D.; Farnum, M. A.; Nelen, M. I. Universal screening methods and applications of ThermoFluor. J. Biomol. Screening 2006, 11 (7), 854−863. (22) Lavinder, J. J.; Hari, S. B.; Sullivan, B. J.; Magliery, T. J. Highthroughput thermal scanning: a general, rapid dye-binding thermal 4930

DOI: 10.1021/acs.jmedchem.7b00208 J. Med. Chem. 2017, 60, 4923−4931

Journal of Medicinal Chemistry

Article

(43) Davis, B. J.; Erlanson, D. A. Learning from our mistakes: The ’unknown knowns’ in fragment screening. Bioorg. Med. Chem. Lett. 2013, 23 (10), 2844−2852. (44) Lea, W. A.; Simeonov, A. Differential scanning fluorometry signatures as indicators of enzyme inhibitor mode of action: case study of glutathione s-transferase. PLoS One 2012, 7 (4), e36219. (45) Alexandrov, A. I.; Mileni, M.; Chien, E. Y. T.; Hanson, M. A.; Stevens, R. C. Microscale fluorescent thermal stability assay for membrane proteins. Structure 2008, 16 (3), 351−359. (46) Krȩzė l, A.; Bal, W. A formula for correlating pKa values determined in D2O and H2O. J. Inorg. Biochem. 2004, 98 (1), 161− 166.

4931

DOI: 10.1021/acs.jmedchem.7b00208 J. Med. Chem. 2017, 60, 4923−4931