Developments with Bead-Based Screening for Novel Drug Discovery

Abstract

Introduction

Combinatorial chemistry provides a cost-effective method for rapid discovery of drug hits/leads. The one-bead-one-compound (OBOC) library method is in principle ideally suited for this application, because it permits a large number of structurally diverse compounds to be rapidly synthesized and simultaneously screened for binding to a target of interest. However, application of OBOC libraries in drug discovery encountered significant technical challenges.

Areas covered

This Special Report covers the challenges associated with first-generation OBOC libraries (difficulty in structural identification of non-peptidic hits, screening biases and high false positive rates, and poor scalability). It also covers the many strategies developed over the past two decades to overcome these challenges.

Expert opinion

With most of the technical challenges now overcome and the advent of powerful intracellular delivery technologies, OBOC libraries of metabolically stable and conformationally rigidified molecules (macrocyclic peptides and peptidomimetics, rigidified acyclic oligomers, and D-peptides) can be routinely synthesized and screened to discover initial hits against previously undruggable targets such as intracellular protein-protein interactions. On the other hand, further developments are still needed to expand the utility of the OBOC method to non-peptidic chemical scaffolds.

1. Introduction

Modern drug discovery often starts with high-throughput screening (HTS) of compound libraries against disease-relevant molecular targets, which requires large collections of high-quality compounds, sophisticated equipment, and significant resources. As such, HTS is generally off limits to most of the academic laboratories and small biotech companies. Over the past three decades, the need for more cost-effective solutions to initial hit identification has led to the development of a variety of combinatorial methods capable of rapidly synthesizing and screening compound libraries of enormous structural diversity (up to 10¹⁴ different compounds).¹ One of these methods produces support-bound libraries in the one bead-one compound (OBOC) format, in which each bead carries multiple copies of a unique compound.² OBOC libraries are readily synthesized by the split-and-pool method^2–4 and all library compounds may be simultaneously screened for binding to a target of interest, e.g., by simply incubating the library beads/compounds with the fluorescently labeled target. Positive hits (e.g., fluorescent beads) are identified by viewing the treated beads under a low-power fluorescent microscope and isolated from the library using a micropipette. The OBOC method was initially demonstrated with peptide libraries and the structures of hit peptides were determined by Edman sequencing² or tandem mass spectrometry.^5,6 The OBOC method is compatible with a broad range of building blocks and chemical reactions, and has been applied to generate various peptidomimetic^{6, 7} and non-peptidic small-molecule libraries.^{8, 9}

Today, combinatorial chemistry has become an increasingly important tool in drug discovery, as many pharmaceutical companies are leveraging DNA-encoded libraries^10–14 and/or biologically synthesized compound libraries (e.g., phage¹⁵ and mRNA display^{16, 17}) for identification of novel hits. At least one FDA-approved small-molecule anticancer drug, sorafenib (which is a multikinase inhibitor), has been discovered through screening an OBOC library.¹⁸ Over the years, OBOC libraries have fallen out of favor after an initial period of excitement in the 1990’s. This is because the first generation of OBOC libraries were associated with several key technical challenges, which prevented their vast potential from being fully realized. However, after nearly three decades of persistent efforts from numerous laboratories, many of these technical challenges have been or are being overcome. Meanwhile, the frontier of drug discovery has shifted towards non-traditional, more difficult drug targets such as protein-protein interactions (PPIs). PPI targets generally do not contain well-defined binding pockets and prove challenging for small-molecule drugs. Indeed, HTS of conventional compound libraries against PPI targets has often failed to produce quality hits. There is a growing consensus that PPI targets require a different class of drug modalities, namely larger molecules that engage greater surface areas on the target. These factors have led to a recent renaissance of OBOC libraries. In this Special Report, we provide a brief summary of the key developments in overcoming the challenges associated with first-generation OBOC libraries and their recent applications in the discovery of conformationally rigidified molecules as a novel class of PPI inhibitors.

2. Recent Developments in Technology

The first-generation OBOC libraries, as pioneered by Lam et al. nearly three decades ago,² suffered from several major challenges, including structural determination of non-peptidic hits, high false-positive rates and biased screening results, and poor scalability. To unleash the full power of OBOC libraries, many research groups around the world have devoted considerable efforts to overcome these technical challenges. Below, we briefly discuss some of the most significant improvements to the OBOC technology.

2.1. Structural Identification of Hits

After a positive bead is isolated from an OBOC library, the structure of the hit compound on the bead must be unambiguously determined. For peptide libraries, hit identification was initially carried out by Edman sequencing,² which is expensive and low throughput. More recently, tandem mass spectrometry has become the method of choice.^{5, 19} Another powerful method is peptide ladder sequencing, which involves progressive truncation of a full-length peptide into a series of peptide fragments, either during library synthesis (before library screening)²⁰ or post screening by partial Edman degradation.^{21, 22} Analysis of the fragments by MALDI TOF mass spectrometry produces a family of peaks (a peptide ladder), from which the sequence of the full-length peptide is readily read. These MS-based methods are also effective for peptidomimetic libraries such as β-peptides⁶ and peptoids,^{6, 23} but not for small molecules or cyclic peptides/peptoids. For the latter, different encoding strategies have been developed. For example, Stille and colleagues incorporated a unique set of haloarene tags during the split-and-pool synthesis of each small molecule.²⁴ Post library screening, the haloarene tags were released from the resin and analyzed by gas chromatography to generate a unique set of peaks in the chromatogram (analogous to a fingerprint) and the identity of the small-molecule hit is inferred from the fingerprint. Other types of tags, such as radiofrequency tags,²⁵ were later employed to encode small-molecule libraries. The main limitation of this encoding strategy is that the number of encoded compounds is generally below 10⁵.

To overcome the library size limitation, Lam and colleagues invented spatially segregated beads, which contain functional small molecules on their surface, but linear peptides inside the beads as encoding tags.^{26, 27} When screened against macromolecular targets (e.g., proteins), which are too large to diffuse into the bead interior, only the functional molecules on the bead surface can interact with the target, whereas the encoding peptides do not interfere with library screening. Pei and co-workers applied this strategy to synthesize large libraries of mono-²⁸ and bicyclic peptides²⁹ (up to 30 million different compounds) and screened them to discover specific ligands against PPI targets such as the oncogenic K-Ras G12V mutant.³⁰ Post-screening hit identification was achieved by sequencing the encoding linear peptides by Edman degradation or partial Edman degradation followed by MALDI-TOF mass spectrometry.^{22, 23} Similarly, Kodadek, Lee, and co-workers generated linear as well as cyclic peptidomimetic (e.g., peptoid) libraries by using a similar strategy.^31–33 As an alternative to the encoding strategy, they also prepared cyclic peptide/peptoid libraries containing a cleavable moiety (e.g., a fixed methionine) in the compound structure.^{32, 33} After a positive bead is isolated, the cyclic peptide/peptoid is selectively linearized at the cleavable moiety (e.g., cleavage after a fixed methionine by CNBr) and the latter is sequenced by tandem mass spectrometry.

More recently, Paegel and co-workers introduced DNA as encoders into OBOC libraries, termed “DNA-encoded solid-phase synthesis (DESPS)”, which involves parallel compound synthesis in organic solvent and aqueous enzymatic ligation of unprotected encoding dsDNA during each synthesis step.^{34, 35} Post-screening hit identification is achieved by next-generation sequencing of the encoding DNA. The DESPS technology combines the advantages of solid-phase chemical synthesis and the high sensitivity and capacity of DNA encoding. This in turn permits the use of smaller beads (e.g., 10-μm beads) for OBOC libraries, substantially increasing the library diversity that is experimentally feasible in a typical research laboratory, to ~10¹⁰ different compounds. The smaller beads also have the added advantage of being compatible with standard flow cytometers and positive beads can be isolated from large libraries by flow cytometry (vide infra).

2.2. Screening Biases and False Positives

While on-bead screening is high-throughput and operationally simple, it suffers from high false-positive rates and biased screening results. A major source of biases/false positives is the high ligand loading on conventional OBOC libraries, which permits a macromolecular target to simultaneously interact with two or more ligand molecules on the bead surface. The avidity effect results in a poor correlation between solution-phase binding affinity (i.e., actual binding affinity) and the amount of on-bead binding.³⁶ In extreme cases, it was noted that a protein target with an acidic pI (or an acidic surface patch) bound to any beads carrying peptide ligands that contained one or more positively charged residues (e.g., Arg) through electrostatic interactions, even though the peptides had no significant binding affinity to the protein in solution.³⁷ To alleviate this problem, Lam and Pei groups generated OBOC libraries with reduced ligand densities at the bead surface, but with normal density inside the bead.^{37, 38} It was found that a ≥10-fold reduction in ligand density largely eliminated false positives caused by charge-charge interactions, while the high density inside the beads provided sufficient material for hit identification.³⁷ Other methods that prove effective in reducing false positives include screening redundant libraries and multiple rounds of screening. Kodadek and colleagues found that limiting sequences to those that appeared twice or more during screening greatly reduced the number of false positives.³⁹ Likewise, multi-round screening using different methods can eliminate biases/false positives that are uniquely associated with each method.⁴⁰

Perhaps the most effective (albeit more technically challenging) method to eliminate false positives and screening biases is to follow up on-bead screening with a round of solution-phase screening. Auer and co-workers isolated “positive” beads from a small-molecule OBOC library by the confocal nanoscanning (CONA) method.^41,42 They first employed confocal imaging to differentiate false positive beads due to autofluorescence (which are uniformly fluorescent throughout the bead volume) from those caused by protein binding (which show fluorescence only at the surface). Next, they labeled the hit compounds (while still on beads) with a fluorescent dye molecule, released the labeled compounds from individual beads into solutions, and determined the binding affinity of each compound for the intended protein target by fluorescence polarization. This allowed them to rank order the hits according to the true binding affinities and identify the most potent hit(s). Importantly, this strategy avoids the expensive process of resynthesis and testing of a large number of hits derived from on-bead screening. Other investigators have also interfaced on-bead screening with other secondary assays (e.g., bead array) to identify the most active hits.^43–45

2.3. Improved Screening Platforms and Automation

In the great majority of studies reported so far, OBOC libraries were screened for direct binding to molecular targets. Identification of positive beads often relied on viewing a pool of beads under a light or fluorescence microscope. This is obviously not ideal for drug discovery on an industrial scale. To automate the screening process, Heusermann et al. attached a polymer capillary to a semiautomated bead-picking device, which allows the operator to isolate individual hit beads in less than 20 s.⁴⁶ The system was used to screen >200,000 bead-bound compounds in 1.5 h. Flow cytometers, which were originally developed for cell sorting, have also been adapted for screening OBOC libraries after incubation with fluorescently labeled proteins.⁴⁷ The advent of DNA-encoded libraries, which can be synthesized on 10-μm beads, has made flow cytometry a more popular choice for OBOC library screening, as it can be performed on standard flow cytometers without any modification.^{48, 49} However, these two methods are still relatively slow for screening very large OBOC libraries. Since most flow cytometers are limited to ~10,000 cells/s, screening 1 g of 10-μm OBOC library (2.0 × 10⁹ beads) would take 56 hours!

Finally, magnetic sorting provides a truly high-throughput method for separating hits from non-hits in an OBOC library.^{19, 43, 50} In this case, the OBOC library is incubated with a biotinylated target of interest and streptavidin-coated magnetic particles (e.g., Dynabeads). Binding of the target (and the associated magnetic particles) to positive beads renders the latter magnetic, which are then attracted to the wall of a container by placing a powerful magnet next to the container, while the non-hit beads settle to the bottom of the container and are carefully removed with a pipette. This simple technique is highly effective, capable of screening millions (or even billions) of OBOC beads in less than an hour. A drawback of magnetic sorting is the relatively high false-positive rates, as some non-hit beads tend to stick with hit beads and are also attracted to the wall. Nevertheless, magnetic sorting rapidly reduces the number of library beads from many millions to typically <1000 and the resulting enriched bead population can be further screened by secondary assays. Lawler and colleagues later constructed a microfluidic device to facilitate high-throughput isolation of magnetized positive hit beads from OBOC libraries.⁵¹ They reported that the device was able to sort magnetized beads with superior accuracy compared to conventional manual sorting approaches, thus offering a very convenient yet inexpensive alternative for screening OBOC libraries. As a result of the above optimization and automation efforts, it is now possible for a single researcher to synthesize, screen, and identify the most active compound(s) from a >1 million-member OBOC library in as few as 4 to 5 days, without the use of any special equipment.^{45, 52}

Efforts have also been made to screen OBOC libraries for biological activities (e.g., enzyme inhibition) other than target binding. This generally requires release of library compounds from the solid phase, which would normally result in a mixture of library compounds in the same solution. To overcome this problem, Paegel and co-workers engineered a microfluidic device to generate aqueous droplets that encapsulate OBOC library beads, a molecular target of interest (e.g., an enzyme), and the proper assay reagent (e.g., a fluorogenic substrate) in an oil medium.⁵³ In essence, each droplet acts as an individual reaction vessel and contains a single bead/compound. UV irradiation photochemically releases the compound from the library bead, which then exerts its biological activity (if any) on the target in solution. As an initial proof of principle, the investigators tested pepstatin A for inhibition of HIV-1 protease and found that droplets containing pepstatin A beads exhibited diminished fluorescence, whereas empty droplets (no beads) exhibited the highest fluorescence due to uninhibited proteolysis of a fluorogenic substrate. The investigators later interfaced the microfluidic droplet system with DNA-encoded OBOC libraries.³⁵

3.0. Conclusion

In summary, most of the technical challenges associated with first-generation OBOC libraries have now been resolved. Small molecules and other compounds that cannot be directly sequenced by Edman degradation or tandem mass spectrometry can now be prepared and screened against intended targets by encoding them with linear peptides or DNA tags. Spatially segregated beads may be used to confine the encoding tags to the bead interior, preventing them from interfering with library screening. They also permit the synthesis of OBOC libraries with reduced ligand density at the bead surface, increasing the stringency of library screening while largely eliminating nonspecific binding caused by the avidity effect. Sensitive assays have been developed to quantitatively determine the binding affinity of initial hits to the intended target in the solution phase, by using compounds released from single 90-μm beads. This eliminates the expensive step of individual resynthesis and testing of a large number of initial hits. Finally, a number of automation technologies (e.g., flow cytometry and magnetic sorting) enable ultra-high-throughput screening of millions of OBOC beads/compounds very rapidly and inexpensively. These technological advancements have recently led to a renaissance of OBOC libraries in chemical biology and drug discovery.

4.0. EXPERT OPINION

It is our opinion that the OBOC technology has advanced to a stage when it is ready to have a major impact on drug discovery, especially for the nontraditional drug modalities described below. Compared to phage display, mRNA display, and DNA-encoded libraries, which are already widely used for drug discovery, OBOC libraries are synthesized in organic solvents and compatible with a much broader selection of building blocks (e.g., D- and N^α-methylated amino acids) and chemical reactions. As such, the OBOC method is well suited for generating more complex molecules that cannot be easily prepared by the DNA/RNA-encoded methods. In our opinion, one such class of molecules are macrocycles including mono-^{30, 45} and bicyclic peptides^{29, 54} and cyclic peptidomimetics.^{32, 33} A second class of interesting compounds are acyclic oligomers made of conformationally rigidified building blocks.⁵⁵ A third of class of exciting molecules are D-peptides, which are metabolically stable and essentially non-immunogenic. Because of their larger sizes (relative to traditional small-molecule drugs) and increased conformational rigidity (in the case of first two classes of molecules and relative to linear peptides and peptoids), these molecules are capable of inhibiting challenging drug targets such as PPIs with antibody-like affinity and specificity. Although some of these molecules (e.g., macrocyclic peptides) have been synthesized by the DNA/RNA-encoded methods,⁵⁶ phage and mRNA display are largely restricted to proteinogenic amino acids and certain unnatural L-amino acids, limiting the structural diversity and proteolytic stability of the resulting macrocyclic peptides. Mirror-image phage display has been developed to discover D-peptide ligands,^{57, 58} but the method is only applicable to protein targets that are small enough to be chemically synthesized. On the other hand, DNA-encoded library synthesis^10–14 is compatible with unnatural building blocks including D-amino acids, but is limited to building blocks and synthetic chemistry that are compatible with the DNA structure and the aqueous medium. Another key obstacle that has in the past dampened the drug industry’s interest into the above classes of molecules (especially peptides and peptidomimetics) is the fact that they are generally impermeable to the cell membrane. Fortunately, recent developments in the latter area suggest that all three classes of molecules may be rendered cell-permeable, through passive diffusion or conjugation to cell-penetrating peptides.⁵⁹ These developments make the three classes of molecules exciting modalities for targeting undruggable proteins, such as those involved in intracellular PPIs. We expect the OBOC method to be the method of choice for initial hit discovery from these types of molecules.

OBOC synthesis has thus far been explored for a rather limited number of chemical scaffolds, the majority of which are peptidic or peptide-like. It is important for researchers to continue to push the envelope of the OBOC technology, especially for synthesis and screening of natural product-like small molecules and non-peptidic macrocycles. Unlike peptides and peptidomimetics, which are assembled through repetitive but highly efficient amide bond formation, synthesis of small molecules and non-peptidic macrocycles requires a much more diverse array of chemical reactions. Because OBOC synthesis leaves no opportunity for purification of synthetic intermediates, these reactions must be rendered highly efficient (ideally with quantitative yields) on the solid phase. Because small molecules and macrocycles cannot be directly decoded by tandem mass spectrometry, novel encoding strategies compatible with both diverse chemical reactions and large library sizes (>10⁵) need to be developed. Finally, further improvements of the screening process would be highly desirable, with respect to both stringency (i.e., less false positive and screening biases) and scalability (i.e., automation).

Article Highlights.

• OBOC libraries provide a cost-effective method for discovery of initial hits in drug discovery, but the first-generation libraries encountered significant challenges.

• The challenge in structural identification of non-peptidic hits has been solved by using encoding tags, such as linear peptides and DNA.

• False positives and screening biases have been alleviated by reducing the ligand density on the bead surface and following on-bead screening with additional rounds of different screening assays.

• More scalable screening platforms (e.g., by flow cytometry and magnetic sorting) have been developed.

• Methods have been developed to screen OBOC libraries for biological activities other than direct target binding.