Target Immobilization as a Strategy for NMR-Based Fragment Screening

Abstract

Fragment-based drug discovery (FBDD) has become a widely accepted tool that is complementary to high-throughput screening (HTS) in developing small-molecule inhibitors of pharmaceutical targets. Because a fragment campaign can only be as successful as the hit matter found, it is critical that the first stage of the process be optimized. Here the authors compare the 3 most commonly used methods for hit discovery in FBDD: high concentration screening (HCS), solution ligand-observed nuclear magnetic resonance (NMR), and surface plasmon resonance (SPR). They selected the commonly used saturation transfer difference (STD) NMR spectroscopy and the proprietary target immobilized NMR screening (TINS) as representative of the array of possible NMR methods. Using a target typical of FBDD campaigns, the authors find that HCS and TINS are the most sensitive to weak interactions. They also find a good correlation between TINS and STD for tighter binding ligands, but the ability of STD to detect ligands with affinity weaker than 1 mM K_D is limited. Similarly, they find that SPR detection is most suited to ligands that bind with K_D better than 1 mM. However, the good correlation between SPR and potency in a bioassay makes this a good method for hit validation and characterization studies.

Introduction

Parallel advances in the fields of combinatorial chemistry and robotics in the 1990s led to the adoption of high-throughput screening (HTS) as the dominant technique in the lead generation stage of drug discovery. In HTS, the main objective is to identify the most potent molecules within libraries as large as 10⁶ compounds by screening techniques based on binding, enzymatic, or cellular assays. The assays used typically are limited to detecting actives with potency better than 1 µM. To achieve this potency, compounds must have a sufficient number of appropriate interactions with the target. As a result, HTS libraries contain compounds that are relatively large (250-500 Da) and complex (as defined by Hann et al. ¹ ) in comparison, for example, to those used for fragment-based drug discovery (see below). Although HTS has undoubtedly generated valuable leads, the molecular complexity and weight of HTS hits place constraints on the possible routes to evolve their potency and specificity while retaining characteristics that lead to good ADME properties. ² Furthermore, the complexity of molecules screened in HTS results in low hit rates due to their reduced complementary to the binding site of a protein. ¹ In addition, the fraction of “compound space” that can be represented in HTS screening decks is extremely small. ³ Fragment-based drug discovery (FBDD), in which small, typically 150 to 300 Da, so-called drug fragments are screened for interaction with the target, has gained increasing interest as a response to these issues.

The use of fragments as starting points for drug discovery has a number of advantages. Fragments make better use of both the scaffold and functional groups to bind the target with the result that their ligand efficiency (LE, defined as ΔG_bind/heavy atom count ⁴ ) is significantly higher than that of compounds found by HTS. ⁵ Higher LE allows increased flexibility when elaborating hits and critically allows for a smaller lead compound. ⁶ Furthermore, small-molecule fragments possess better complementarity to target proteins, resulting in a higher hit rate than HTS, ¹ and because the available chemical space is many orders of magnitude smaller, it can be much more efficiently sampled. As a result, FBDD requires a small but rationally designed library that can be carefully controlled to ensure quality. However, fragments, due to the limited potential to productively interact with the protein, typically bind with K_D > 10 µM. Such low-affinity hits can only be identified using highly sensitive biophysical methods such as high-throughput X-ray crystallography, ⁷ nuclear magnetic resonance (NMR), ⁸ and surface plasmon resonance (SPR) ⁹ or, when appropriate, an enzyme inhibition assay performed at high concentration of the fragment. ¹⁰ NMR screening methods range from the detection of signals from ¹⁵N- or ¹³C-labeled target proteins ⁸ to so-called ligand-observed methods, which monitor the ¹H or ¹⁹F spins of the fragments. ^11,12 However, because NMR remains a mass insensitive technique, screening of even relatively small fragment libraries still requires 10s to 100s of mg of purified protein. Recently, we developed a screening method called target immobilized NMR screening (TINS), in which a target and a reference protein are immobilized on solid media, and binding of fragments to the immobilized proteins is monitored by NMR. ¹³ One advantage of this method over others is the small amount of protein required because TINS uses a single sample of the target to screen an entire fragment library. ¹⁴

Because fragments are by definition small, even those that bind the target very weakly may be characterized by high LE. For example, a fragment containing 12 heavy atoms with a K_D of 2 mM has an LE > 0.31, a generally accepted threshold for a desirable starting point for elaboration. ⁴ Therefore, the ligand screening technique must be capable of detecting weak binding well into the single-digit mM range. On the other hand, the technique should be robust to avoid a high false-positive rate. Here we present a comparison study of TINS to the fragment screening techniques most commonly used in FBDD—namely, saturation transfer (STD) difference NMR spectroscopy, ¹¹ SPR, and high-concentration bioassay. The viral RNA-dependent RNA Polymerase (RdRP) was selected as a target because it is both typical of an FBDD project and amenable to all assay formats. ¹⁵ A fragment library was initially screened against RdRP using TINS. Based on this screen, 133 fragments were selected, including both hits and nonhits, and rescreened using STD. The hits generated by both methods were characterized in a bioassay for inhibition of RdRP activity. A subset of fragments that were positive in all 3 assays was then characterized for binding to RdRP using an SPR biosensor assay. On the basis of these results, we provide recommendations for the optimal use of each method.

Materials and Methods

Fragment preparation for TINS and 1D-STD

The composition of the fragment library has been described. ¹⁶ The molecular weight of the fragment library ranged from 150 to 300 Da and averaged 220 Da. Compounds were dissolved in DMSO-d₆ at 50 or 100 mM and were subsequently diluted to 500 µM in the final aqueous buffer solution used for screening. The fragment mixes consisted of typically 3 to 5 fragments. For both STD and TINS screening, fragment mixes were prepared in screening buffer, which comprised 25 mM Tris-d₁₁ (pH 7.5), 150 mM NaCl, and 1 mM MgCl₂ in 99.9% D₂O.

Target preparations

The viral RdRP, which included a C-terminal 6 His-tag (63 kDa), was expressed in Escherichia coli strain BL21 DE3+ and purified using a combination of nickel affinity and cation exchange chromatography. The C-terminally 6 His-tagged PH domain (aa 1-123) from human Akt1 (15 kDa) was expressed using BL21 DE3+ cells and purified using a similar procedure for RdRP.

High-concentration bioassay

Polymerase activity was assayed as described. ¹⁵ The percentage of inhibition (PIN) per reaction was calculated as follows:

$$\text{PIN}(\%)=[1-({\text{A}}_{\text{fragment}}-{\text{A}}_{\text{positive control}})/({\text{A}}_{\text{negative control}}-{\text{A}}_{\text{positive control}})]\times 100\%,$$

where A_fragment and A_{negative control} are the radioactivity counts measured in the presence of the fragment and DMSO, respectively, whereas A_{positive control} is the radioactivity measured in the presence of a known inhibitor that abrogates enzymatic activity completely. The assay was performed in duplicate for samples and in triplicate for the negative (0.5% or 2% DMSO) control and the positive control in the presence of a known inhibitor at 250 µM. The quality of the assay was monitored by Z′ score calculated from the controls. ¹⁷

TINS methodology and analysis

In total, 4 mg of the viral protein and 1 mg of Akt1 PH domain were immobilized via amine coupling to 500 µL (bed volume) of Actigel-ALD resin (Sterogene, Carlsbad, CA) according to the manufacturer’s protocol in 50 mM Na phosphate (pH 7.4) and 150 mM NaCl solution overnight at 4°C. Typically, an immobilization efficiency of 90% was achieved using this protocol, and the final concentration of immobilized protein was estimated to be in the range of ~110 µM. Unreacted sites on the resin were blocked by addition of 100 mM d₁₁-Tris. The resin was then packed into a dual-cell sample holder. ¹⁴

TINS NMR measurements were performed as described ¹⁴ with the exception that a spatially selective Hadamard pulse sequence ¹⁸ was used. A T2 delay of 80 ms using a CPMG sequence was added prior to signal acquisition to suppress resonances arising from the resin. The change in signal amplitude due to interaction of the fragment with either target or reference protein is referred to as the TINS effect and was calculated as a weighted average of the TINS effect observed for each unique signal (n) of a fragment:

$$\text{TINS effect}={\displaystyle {\sum }_{\text{n}}(}1-{\text{H}}_{\text{target}}/{\text{H}}_{\text{reference}})/\text{n},$$

where H is the height of the observable signals. The amount of solution in the target and reference cells can be determined by injecting the TINS running buffer in H₂O into each cell and integrating the signal. In general, the difference in the integrated volume of water signal in the 2 cells must be below 10%. This difference is reflected in the difference of the height of resonances of nonbinding reference molecules (0.5 mM imidazole, 0.5 mM Na acetate, 4% d6-DMSO, and 0.5 mM TMA) between the 2 cells. Therefore, 0.1 was used as the lowest cutoff value for a significant TINS effect in the hit identification process.

STD methodology and analysis

All STD NMR measurements were performed using a Bruker AV 500-MHz spectrometer (Bruker, Billerica, MA) equipped with a TXI probe. The purified viral protein was used at 20 µM, and the fragment was assessed using the same mix and conditions as used in the TINS experiment. The STD spectrum was recorded using the pulse scheme reported previously. ¹¹ The total number of scans was 512, resulting in a 54-min experiment time per sample. The STD effect was calculated as follows:

$$\text{STD effect}={\displaystyle {\sum }_{\text{n}}(({\text{I}}_0-{\text{I}}_{\text{sat}}})/{\text{I}}_0)/\text{n},$$

which expresses the signal intensity (I_sat) in the on-resonance spectrum as a fraction of the intensity of the unsaturated off- resonance spectrum averaged over all observable signals. A cutoff of 0.02 (2 times the standard deviation) for a significant STD effect was used for hit identification in the screen.

SPR biosensor methodology

The interaction between the viral enzyme and validated fragment hits was studied on a Biacore T100 instrument (GE Healthcare, Piscataway, NJ), equilibrated at 25°C. CM5 sensor chips (research grade; GE Healthcare) were used for the experiments. Viral enzyme sensor surfaces and reference surfaces were prepared by standard amine coupling via exposed primary amines (amine coupling kit; GE Healthcare). Typical immobilization levels ranged from 9500 to 12,500 resonance units (RU) and 1800 to 2200 RU, respectively. Surfaces were stable for at least 4 days, as measured by R_max and the signal-to-noise ratio.

For interaction studies between the immobilized viral protein and the fragments, 25 mM Tris (pH 7.5), 150 mM NaCl, 1 mM MgCl₂, 0.005% of the detergent P20 (GE Healthcare), and 0.1% or 0.4% DMSO were used as running buffer. Stock solutions of the fragments were made in DMSO (10 mM). Samples were injected for 60 s with a flow rate of 30 µL/min in a series of 4 concentrations with a maximum of 200 µM. To determine specific binding, the reference channel, containing the immobilized Akt1 PH domain, was subtracted from the experimental channel.

Results

Screen of the fragment library for binding to RdRP using TINS

The viral and reference proteins were immobilized via primary amines, and the functionality of immobilized RdRP was determined using the bioassay described. Subsequently, the immobilized proteins were packed into a dual-cell sample holder. ¹⁴ Due to susceptibility mismatch between the sepharose beads and surrounding aqueous buffer, the line width of small molecules increases from about 1 to 11 Hz when recorded at 500 MHz. This line broadening is specifically taken into account when designing mixtures of fragments so that every fragment has at least 3 well-resolved resonances in the spectrum. Binding of a fragment to a protein molecule immobilized on a solid support causes the resonances of a fragment to further broaden into the kHz range. As a result, ligand binding is detected by a simple reduction in the height of all NMR signals from that fragment ( Fig. 1 ). We tested the specificity of TINS using a number of known ligands of RdRP. Figure 1A demonstrates preferential binding of 4 such known ligands to RdRP over the reference protein, the PH domain of human Akt1. This particular protein was chosen as it presents a typical protein surface yet is devoid of specific small-molecule binding sites. ¹⁹ Use of a reference protein in TINS eliminates very weak, nonspecific binders that might otherwise lead to false positives in the hit identification process. To monitor the integrity of the immobilized RdRP during the screen, binding of several of the known ligands was repeatedly assayed (data not shown). Figure 1B shows binding of 2 compounds in a typical mix of fragments.

Fig. 1.

Screening of the fragment library for binding to RdRP using target immobilized NMR screening (TINS). (A) Binding of 4 different known ligands of RdRP. Pairs of overlaid spectra are shown for each known ligand. In each case, the spectrum with the greater signal intensity is the spectrum of the compound in the presence of the reference protein, the PH domain of Akt1. The asterisk indicates the signal from residual ¹H DMSO (also in part B). (B) Detection of ligand binding in mixes of fragments. The 2 overlaid spectra at the bottom of the panel are respectively ¹H spectra of a mix of 4 fragments in the presence of the reference protein (larger signals) and the target RdRP. A reference spectrum of each compound is provided (upper traces) where it can readily be seen that the compounds represented by the top 2 spectra bind and those by the lower 2 spectra do not bind. The broad resonances between 3.4 and 3.7 ppm derive from the sepharose resin used to immobilize the proteins.

A library consisting of 1270 commercially available fragments was screened against the viral RdRP using TINS. The physicochemical properties of the fragment library have been reported previously. ¹⁶ The library was screened by repeated cycles of injection of mixtures of fragments, which typically consisted of 3 to 5 compounds, into both cells packed with either immobilized target or reference protein. After the mixture of compounds was injected, the flow was stopped, the NMR data were acquired, and then the fragments were washed from the cell prior to application of the next mix. Thus, a single sample of only 4 mg of viral polymerase was required for the entire fragment library screen by TINS. Based on the initial TINS screening data, a total of 30 mixtures, which consisted of 7 mixtures without binders and 23 mixtures containing at least 1 binder per mix (and many specifically selected to have more than 1), were chosen for screening by STD. The 30 mixtures contained 133 unique fragments, from which TINS had identified 74 hits using the criteria given.

1D-STD screen setup

In STD spectroscopy, one observes a difference in the amplitude of ligand resonances in 2 separate experiments, in which the magnetization of the target protein is alternately saturated (on-resonance spectrum) or unsaturated (off-resonance spectrum). ¹¹ The attenuation of the NMR signals of a ligand occurs in the on-resonance spectrum via through-space, noncoherent transfer of saturation from the ¹Hs of the target protein to those of the bound ligand. When a mixture of compounds is assayed in the presence of a target protein, the resulting difference spectrum between the on- and off-resonance saturation spectra will contain only the signals from the ligand, whereas the signals from the nonbinders are removed in the difference spectrum. The sensitivity of the STD experiment on our 500-MHz magnet equipped with a room temperature probe was optimized using a sample of human serum albumin (HSA; 66 kDa) and known binders, including acetaminophen with a K_D of 690 µM, typical for fragments binding to a target. ²⁰ Titration of HSA into 500 µM acetaminophen indicates that the size of the STD effect is linear with the protein concentration range tested ( Fig. 2A ). As a control, STD spectra of each of the 30 fragment mixtures were recorded in the absence of protein to check for artifactual STD signals arising from accidental saturation of fragment ¹H resonances. With the current setup, such artifact STD effects were not observed (data not shown).

Fig. 2.

Saturation transfer difference (STD) experiment was optimized using human serum albumin (HSA) and 500 µM of acetaminophen as a control. (A) Acetaminophen is reported to have a weak affinity of 690 µM to HSA and expected to be similar to affinities of the fragment compound and therefore was chosen for the case study. Only in the presence of HSA (top traces) do STD signals, due to interactions between acetaminophen and the protein, become visible. Linear relationship between the fractional STD signal and the protein concentration can be seen. (B) The signal-to-noise (S/N) ratio of STD signals from a fragment compound of 211 Da binding to the viral protein clearly improved by increasing protein concentration. The improvements in S/N for those 2 signals (around 7.5 ppm) were 3.2- and 4-fold in 20 µM protein sample over 5 µM.

Initially, the STD NMR screening was conducted at 5 µM RdRP, which resulted in far fewer hits than when the screen was performed at 20 µM. Clear improvements in the signal-to-noise (S/N) ratio of STD signals from a binding fragment were observed when the viral protein concentration was raised by 4-fold ( Fig. 2B ). Because screening of fragment libraries using STD is typically performed on an instrument using a cryoprobe, which affords an improvement in S/N of about 3, we screened the 30 fragment mixtures using an equivalently higher protein concentration of 20 µM. This screen required approximately 20 mg of viral protein and generated 49 hits. The amount of protein required could have been reduced to 5 to 7 mg of protein using a cryoprobe.

Comparison of TINS and STD in fragment screening

Using the criteria established, we detected 49 hits from the 133 fragments screened using STD compared to the 74 we had previously detected using TINS. Combining the unique hits from both TINS and STD gives a total of 83 hits. Examples of shared and unique hits from an identical mixture are depicted in Figure 3 (see also below). The robustness of screening in mixtures for each method was assessed by rescreening 29 randomly selected fragments from the mixtures as singletons ( Figs. 1B, C ). The reproducibility of results in mixtures versus singletons was fair for STD (R ² = 0.885) and reasonable for TINS (R ² = 0.948). Based on the data of Figure 3 , screening in mixtures as opposed to singletons would result in missing 1 hit using the TINS approach and 6 hits using STD.

Fig. 3.

Comparison of hits identified by saturation transfer difference (STD) and target immobilized NMR screening (TINS) from a mix of fragments. (A) A mix consisting of 5 fragments was assessed for target-specific binding using TINS (upper spectra) and STD (lower spectrum). The fragments that were identified as hits using the cut-off criteria (see methods) are numbered. The compound labeled X clearly does not bind either to the target or reference as its TINS spectra are identical in both. In contrast, fragments 1 to 3 bind specifically to the target, as evidenced by a similar reduction in the height of all signals from each compound. The spectral region (3.4-5.5 ppm) indicated by the asterisk is not included in the TINS analysis as the region contains residual signals from the resin as well as artifacts arising from solvent suppression. Reproducibility of (B) 1D-STD and (C) TINS effects when screened as singletons or in mixes. Twenty-nine fragments from the mixes were reassayed at 500 µM each as a singleton, and the STD and TINS effects were compared to the corresponding effects observed when screened in mixes. The dotted lines are the linear fit of the 29 points.

Characterization of NMR hits by a high-concentration bioassay

High-concentration bioassays have been successfully applied to the initial stage of fragment hit identification. ²¹ Typically, biophysical methods (e.g., X-ray, NMR, and SPR) are used to validate bioassay hits by monitoring the physical interaction with the target. In the present study, a high-concentration bioassay was performed on the NMR hits to characterize their inhibitory activity against the RdRP polymerase. The 83 total hits were individually assayed at 500 µM in the polymerase assay. The criteria applied in classifying a fragment as an inhibitor vary significantly. In looking at only 2 examples, one finds thresholds of (using PIN) PIN >30% at the 200-µM fragment ²² and a PIN >60% at the 1.0-mM fragment. ²³ Here, fragments were considered “inhibitors” when they exhibited a PIN value above 20% (group I in Fig. 3A ), which was the 3 times the average standard deviation (ASD) of the entire enzymatic assay. The fragments that exhibited a PIN value ≥−10% and <20% at 500 µM were further assayed at a 2-mM concentration to maximize identification of even weaker inhibitors. For the 33 fragments reassayed at 2 mM, the 22 with PIN values above 30% (= 3 × ASD) were classified as “weak inhibitors” (group II) and are indicated by filled squares in Figure 3A . The 11 fragments that exhibit no significant inhibition at either concentration were classified as biologically inactive (group III). Some of the fragments reproducibly stimulated the activity of RdRP (group IV), but because both TINS and STD only detect binding, we classified this latter group of fragments as validly interacting with the target and demonstrating biological activity. In total, 70 of the 83 NMR-defined hits showed significant activity in the biological assay—that is, they were members of group I, II, or IV. Here actives could include reversible, stoichiometric, competitive or noncompetitive (a “genuine positive”), and destabilization or precipitation of the enzyme (a “false positive”).

Comparison of NMR to high-concentration bioassay for ligand identification

An inherent advantage of a high-concentration screening is that it can be readily quantified. We wished to assess whether either the TINS or STD effects observed correlated with the PIN at the given fragment concentration. For this purpose, we treated an activator as equivalent to an equipotent inhibitor. As shown in Figures 4B,C , there is a weak correlation between the absolute value of the PIN and the STD effect (R ² = 0.68) and no correlation between the PIN and the TINS effect. This means that the size of the STD effect, as assayed for library screening, only weakly translates into the bioassay activity of a fragment hit, whereas that of TINS does not at all.

Fig. 4.

Comparison of saturation transfer difference (STD) and target immobilized NMR screening (TINS) effects to biological activity. (A) The percentage of inhibition (PIN) was determined for each fragment defined as a hit in either the TINS or STD screen at 500 µM in a polymerase inhibition assay (see Materials and Methods). The PIN of each fragment is displayed. Group I (filled diamonds) is fragments with PIN >20%. Group II (open squares) had PIN values −10% < PIN < 20% at 500 µM and PIN >30% at 2 mM. Group III fragments (filled triangles) were those with PIN <30% at 2 mM. Fragments that reproducibly stimulated the enzyme were designated group IV (open circles). Comparison of (B) PIN to STD effect or (C) TINS effect. The PIN was plotted against the corresponding measure of binding from the (B) STD screen or (C) TINS screen. In both cases, there is no correlation between the size of PIN and the nuclear magnetic resonance (NMR) effect.

A summary of results from the 3 approaches is presented in Table 1 . Our results clearly indicate that both STD and TINS are readily capable of identifying the most potent fragments, although TINS identified twice the number of unique hits in comparison to STD. Examples of fragments that generated a positive response in all 3 assays or either one of the NMR assays and the enzyme inhibition assay are provided in the first 3 rows of Table 2 . Interestingly, the biochemical assay picked up 2 hits that were negative in both TINS and STD, possibly suggesting an artifactual mode of inhibition ( Table 2 , row 5). However, a significant difference in the ability to detect weakly binding ligands is observed where TINS finds 2.5 times the number of fragments that are biologically active as STD. Of the 9 TINS positives that were biochemically negative, 3 were also positive in STD ( Table 2 , row 6). These data suggest that either the compounds bind too weakly to inhibit or they bind at a nonbiologically productive site.

Table 1.

Comparison of Hits from STD, TINS, and Bioassay

Method	Total Hits	Group I Inhibitors	Group 2 Bioassay at 2 mM PIN >30%	Group 3 Bioassay at 2 mM PIN <30%	Group IV Activator	Group V NMR Positive Bioassay Negative
STD	49	30(4)	7	3	5 (2)	—
TINS	74	33(8)	19	9	6 (4)	—
Bioassay	70	40	22	9	8	3

STD, saturation transfer difference; TINS, target immobilized NMR screening; PIN, percentage of inhibition; NMR, nuclear magnetic resonance.

Table 2.

Examples of Compounds with Results of Each Detection Method

Row	TINS	STD	EI	SPR
1	+	+	+	+
2	+	−	+	+
3	−	+	+	+
4	−	−	+	+
5	−	−	+	−
6	+	+	−	−
7	+	+	+	−

EI, enzyme inhibition; STD, saturation transfer difference; TINS, target immobilized NMR screening; SPR, surface plasmon resonance.

Fragment screen by SPR biosensor

We wished to compare the ligand screening results of the NMR and biochemical methods to those obtainable by SPR using a Biacore T100 (Biacore, Piscataway, NJ), an instrument typically employed for this purpose. As with TINS, the target is immobilized to detect binding in SPR. Subsequently, a solution of the fragment to be tested (one at a time) is flowed over the surface of immobilized target, and the change in signal detected upon binding is in general proportional to the mass of the compound and the number of available binding sites on the surface. Because fragments have a low molecular weight, high surface densities of immobilized protein are necessary. ⁹ A total of 62 fragments from the STD and TINS hits and 1 known ligand with a mass of 385 Da (IC₅₀ of ~10 µM) were assayed for binding to RdRP using SPR with the Akt1 PH domain as a reference. The selected fragments consisted of 30 from group I, 23 from group II, 7 from group III, and 2 fragments that were negative in both STD and TINS but positive in the enzyme assay.

To establish the validity of our SPR assay, we first titrated the known ligand at a concentration range of 1.25 to 10 µM to establish a linear response ( Fig. 5A ). Although the experiments were carried out in a manner most appropriate for ligand screening, we could estimate an approximate affinity for some of the compounds by fitting the concentration series. Using this approach, the K_D of the known ligand was estimated to be 5 µM, which is consistent with the known IC₅₀ of 10 µM. Subsequently, the 62 fragments were titrated at the following concentrations: 6.25, 12.5, 25, and 50 µM. Fragments were identified as hits when the signal at each of the concentrations exceeded 3 times the noise and a clear concentration-dependent response was observed. Examples of fragments from this class are shown in Figure 5B and Table 2 , row 1. Using these criteria, 16 of the 62 fragments were considered hits in the first concentration series. Of the 16 fragments, 14 were from group I and 2 were from group IV, the enzyme activators. For 8 of these fragments, the data were of sufficient quality to estimate K_Ds, and these ranged between 10 and 600 µM. This group was designated high-affinity ligands in the SPR assay. The remaining 46 fragments were reassayed using a higher concentration range (25-200 µM). Fragments were identified as hits when both the following were true: the binding signal at each concentration exceeded 2 times the noise and a clear concentration-dependent response was observed. Using these criteria, 13 additional fragments were considered hits. One of the members of this latter group included one of the fragments that was negative in both STD and TINS but positive in the bioassay ( Fig. 5C and Table 2 , row 4). This compound was deemed a false negative for both STD and TINS. The K_Ds could be estimated for only 2 of these 13 fragments, and they were in the high µM to low mM range. Accordingly, we designated this group of ligands low-affinity fragments. Binding of the remaining 32 fragments could not be detected using SPR, even at concentrations up to 200 µM. A typical example of the sensorgram for this latter group is shown in Figure 4D ( Table 2 , row 7). Interestingly, 2 of the fragments showed both slow on and off rates, as exemplified in Figure 4E . These 2 fragments were positive in both STD and TINS, although they were near the cutoff for detection in both of these biophysical assays despite the fact that they were among the more potent inhibitors (group I).

Fig. 5.

Example sensorgrams for the interaction between RdRP and test compounds. Each sensorgram depicts binding data for a concentration series of that compound. (A) A known ligand of 385 Da (IC₅₀ ~10 µM) measured at the following concentrations: 1.25, 2.5, 5, and 10 µM. (B) A 279-Da fragment from group I measured at the following concentrations: 6.25, 12.5, 25, and 50 µM. (C) A 148-Da fragment from group I that was negative in both saturation transfer difference (STD) and target immobilized NMR screening (TINS) measured at the following concentrations: 25, 50, 100, and 200 µM. (D) A 211-Da fragment from group 1 that showed no binding when measured at the following concentrations: 25, 50, 100, and 200 µM. (E) A 215-Da fragment that showed unusually slow kinetics measured at the following concentrations: 6.25, 12.5, 25, and 50 µM.

We compared the results of the SPR assay to the biochemical assay by grouping the SPR results in compounds that were “not detected,” “low affinity,” “high affinity,” and “positive control” according to the definitions above and plotted these against the observed inhibition in the bioassay ( Fig. 6 ). Of the 30 group I fragments assayed, 14 (47%) were positive in the SPR concentration series up to 50 µM, whereas another 8 (27%) were positive in the concentration series up to 200 µM. The remaining 8 group I fragments (27%) could not be detected with SPR; an example of these is given in Table 2 , row 7. Two group II compounds were positive in the higher concentration assay (out of 23 assayed), whereas no group III compounds were positive in the SPR experiment. In total, 3 group IV compounds (activators) were positive in SPR, 2 in the high-affinity group and 1 in the low-affinity group. Two compounds were positive in both STD and TINS but negative in the enzyme assay and SPR ( Table 2 , row 6). As readily seen in Figure 6 , there is a quite reasonable correlation between the size of the PIN and the binding response in SPR.

Fig. 6.

Comparison of the interaction between fragments and RdRP as measured by surface plasmon resonance (SPR) to the percentage of inhibition (PIN) obtained from the bioassay. Interactions on SPR are shown on the x-axis and divided into fragments that did not bind in the SPR assay, low and high affinity measured on SPR and the positive control. Red symbols indicate fragments from group II.

Discussion

In this present study, we compared the sensitivity and robustness of 4 different methods for primary screening of fragment libraries for binding to a pharmaceutical target: 2 different ligand-observed NMR techniques, STD and TINS; high-concentration bioassays; and SPR. There are a number of characteristics that a screening technique must encompass to be of use in FBDD. A technique should be broadly applicable to (nearly) all potential targets and sensitive to a weak binding well into the mM range while generating a minimum of false positives. Sensitivity to mM binding is important as small fragments may still exhibit good LE despite a K_D of 2 to 3 mM, and certain target classes, such as protein-protein interaction targets, ²⁴ inherently bind fragments more weakly. Moreover, even if a fragment binds with K_D in the range of 10 mM, we have readily found close analogs with K_D 20-fold tighter, which could result in an increase of the LE by 0.08 to 0.1. Below we discuss each of the 4 methods in light of these characteristics.

Ligand-observed NMR and STD, as first proposed by Mayer and Meyer, ¹¹ and a similar method known as waterLOGSY ²⁵ are widely used NMR techniques for detecting fragment binding. STD and waterLOGSY have the advantage of being relatively easy to implement, requiring only standard NMR instrumentation (optimally with a cryoprobe head) and alleviating the need for an isotopically labeled target, while reducing the amount of protein required by a factor of 10. On the other hand, the sensitivity of STD is dependent on the efficiency of saturation transfer, which is a function of the correlation time τ_c of the target. Because τ_c generally scales with the size of a molecule, both STD and waterLOGSY become more efficient with increasing molecular weight of the target protein, and therefore neither method is optimal for proteins less than about 20 kDa. Furthermore, the protein must remain soluble even in the presence of high concentrations of fragments in low salt (less than 50 mM total ionic strength) to take advantage of the increased sensitivity afforded by a cryoprobe. If the latest generation cryoprobe with a sample volume of 30 µL is used, then 1500 compounds can be screened using about 15 mg of a target the size of RdRP. However, more typically, a 5-mm probe is available, requiring a reduction of the protein concentration of 5 µM, requiring about 50 mg of protein. In this case, the real limitation of solution ligand-observed NMR methods is the lack of sensitivity to weak binding in the mM range.

The use of TINS as a primary screening technique has a number of potential advantages. Our results clearly indicate that it is most sensitive to weak binding. For example, in one project, a ligand with a K_D of 8.9 mM was found that, due to Mr 135 Da, had a LE of 0.28 (to be published elsewhere). Subsequently, a crystal structure of the target-fragment complex was solved, enabling rational elaboration. TINS also makes efficient use of the target, requiring only 4 mg of RdRP to screen the entire fragment library while ligand binding and enzymatic activity of targets routinely is unchanged even after screening 1500 fragments. ²⁶ The use of a reference reduces the number of false positives and, by careful selection of the reference, allows one to adjust the scope of a screen to, for example, focus on one binding site or select fragments specific for one member of a protein family. Using the PH domain of human Akt1, we routinely observe 7 to 10 compounds per screen that preferentially bind the reference, as compared to 30 to 100 that preferentially bind the target. Last, TINS is applicable to a wide variety of targets, more than 20 different proteins, including membrane proteins that have been screened.

We noted a distinct lack of correlation between the binding signal for TINS and the potency in the biological activity, whereas STD fares only marginally better. The poor correlation between the level of biological activity and the magnitude of the NMR readout is likely due to the dependence of ligand-observed NMR signals on the off rate of the ligand. During the fixed length of an NMR experiment, tighter binding ligands, which have slower off rates, will not generate as much signal amplification as weaker binding ligands. This is the likely explanation for the observation that the 2 fragments with slow kinetics in SPR and potent biological activity were only marginally detected as hits in both TINS and SPR. A further issue for TINS is that for ligand screening, the parameters of the NMR experiment are optimized to detect weak binders. By shortening the T2 period during the spatially selective TINS experiment, the correlation between TINS effect and binding affinity can be improved considerably. ²⁷ However, because TINS depends on enhanced transverse relaxation of ligands in the bound state to generate the binding signal, significant inherent differences in R2 between fragments render precise quantitation difficult.

Because the ultimate goal of a small-molecule drug is to modulate the biological activity of a target, using such an assay in the initial stage of drug discovery is logical. This approach has the advantage that it is inherently quantitative. However, high-concentration screening is not a generally applicable approach. First, not all targets have a biological activity than can be readily assayed in vitro, and cell-based assays do not typically possess sufficient sensitivity to detect weakly active fragments. Even when an efficient in vitro biological assay can be developed, it may not be appropriate for screening fragments. For example, protein-protein interactions can often be readily measured using a fluorescently labeled binding partner. However, the low nM affinities characteristic of these interactions are often too tight to be displaced by fragments that bind in the mM range. Last, the high concentration of fragments necessary to generate a response sufficiently above the noise level in a biological assay may lead to artifactual mechanisms of inhibition such as aggregation. ²⁸ However, the combination of a biological assay with a high-quality biophysical assay allows one to eliminate false positives that act through aggregation or nonspecific destabilization mechanisms. Thus, we view biochemical assays, when possible, as an ideal hit validation and prioritization tool after a biophysical assay.

Calculations based on the fundamental principles of SPR suggest that for a ligand of 150 Da, the concentration of the ligand in the SPR assay must be between 0.5 and 1.2 times the K_D for targets in the typical molecular mass range of FBDD. ²⁹ It is unusual to screen fragments for binding at concentrations higher than 500 µM⁹ in an SPR assay, which places an approximate upper limit of detection of 1 mM K_D for moderate-sized targets. These calculations are borne out by our experiments where 82% of the group I fragments, which have sub-mM K_Ds, were detected by SPR, but only a small number of group II fragments, which include many compounds with K_D > 1 mM, could be detected. Therefore, our data suggest that given the present sensitivity of SPR instruments, use of this method as a primary screening approach may lead to an unacceptably high level of false negatives. However, the highly quantitative nature of SPR makes this an extremely powerful, orthogonal technique to validate and prioritize hits from the primary screen. In particular, the combination of TINS for primary fragment screening, in which the format lends itself to ready confirmation of biological activity, and SPR for validation/prioritization should prove a powerful combination to rapidly find and evaluate weakly binding fragment hits for structural studies and elaboration projects.