Chapter 1: Affinity-based DEL Selections with Different Target Types: Overview and Achievements

DNA-encoded library (DEL) technology has witnessed significant advancements since its conceptualization in 1992. ^1–3  DELs are mixtures of small molecules conjugated to DNA tags, whereby small molecules are synthesized on DNA using sequential chemical reactions, each of which is coupled to ligation of an oligonucleotide tag. The DNA sequence functions as the identifier of the chemical structure of each compound in the library, thus facilitating the screening of a vast number of molecules in a mixture of library compounds. The development of combinational chemistry and next generation sequencing (NGS) has enabled the rapid investigation of DELs ranging in size from approximately a million to over a trillion compounds, ^4–6  with a turnaround time as short as a month. Major pharmaceutical companies including GlaxoSmithKline (GSK), Roche, Amgen, Pfizer, Merck (MSD) and others are actively engaged in drug discovery endeavours based on DEL technology. Additionally, DEL service providers such as HitGen, WuXi AppTec and X-Chem have made this technology available to other pharmaceutical companies, smaller biotechs, and academic scientists. With the progression of this technology over the past three decades, DEL chemistry, ^7–9  the copy number of DEL molecules, ^10–12  selection conditions and data analysis ^13–19  have been intensively explored and improved upon. DEL selection now provides an efficient, ultra-high throughput, and cost-effective screening strategy that has made significant achievements and yielded several small molecules that have progressed into different stages of clinical development (Figure 1.1).

Binding processes serve as the basis of all molecules that execute biological functions. DEL selection leverages this principle to investigate affinity-based binding of DEL molecules to their respective targets. Binders selected from DELs are identified through decoding of DNA sequences, and subsequent off-DNA synthesis and biological testing is conducted for hit validation. Different selection strategies have been developed over the past three decades, primarily distinguished by the methods of displaying the target and of partitioning binders from non-binders.

The majority of DEL selection processes employ a solid support such as magnetic beads or resin (e.g. PhyTip columns) to immobilize the target and facilitate the separation of the binders from the bulk DEL solution. The screening target is usually produced with an affinity tag such as hexa-histidine or biotin, to enable immobilization on the solid support (Figure 1.2A). The procedure encompasses several steps including target immobilization, DEL incubation with the protein or nucleic acid target, washing to eliminate non-specific or weak binders, binder release through heating, competitive ligand elution or target elution, and the utilization of the output as the input for subsequent selection rounds. The order of target immobilization and DEL incubation steps can differ in different selection modes, but they all utilize immobilization for DEL separation. ^27,28  DEL selection on a solid support is primarily used due to its straightforward protocol and rapid adaptation to benchtop instruments. All the aforementioned clinical-stage drug candidates were discovered utilizing this selection strategy. Covalent-based DEL selection ^29,30  operates in a similar format but binders are not released given the covalent attachment of the DEL molecule to the target protein, and covalent selections require only one selection round.

Solution-based DEL selection allows investigation of a target in its free state and eliminates the need for target immobilization. Various approaches utilizing different designs have been reported. The binder trap enrichment (BTE) strategy involves the integration of double-stranded DNA to the target protein, enabling the differentiation of bound DEL molecules through selective ligation of the DEL DNA to a DNA tag on the target protein within emulsion droplets. ³¹  The DNA tag on the target protein contains the primer binding site such that ligated constructs are selectively amplified in a subsequent PCR. Similarly, a single stranded DNA tag can be introduced to the target protein and the bound DEL molecule (as single stranded DNA) can be differentiated by complementary recognition and interaction dependent PCR (IDPCR). ^32,33  Another variant of solution-based DEL selection is accomplished by utilizing a fusion protein on the target which can biotinylate or add poly A to the bound DEL molecules by induced proximity. Consequently, the DEL ligands can be enriched with corresponding streptavidin ³⁴  or poly(dT)25 beads. ³⁵  Solution-based selections can also be achieved by using a DNA-encoded dynamic library (DEDL) and PCR of bound DEL molecules upon target binding and photo-crosslinking (Figure 1.2B). ³⁶  These methodologies alleviate the concern of a protein losing activity after immobilization, but specific instrumentation, DEL design, and customized selection methods may be required.

Separation methods based on capillary electrophoresis (CE) and liquid chromatography (LC) have been reported. A CE-based selection relies on the different electrophoretic rates of protein, free DEL, and the protein–DEL complex, ^37,38  while an LC-based approach immobilizes protein on the resin and packs it into the column as a stationary phase (Figure 1.2C). ^39–41  The chromatography-based strategies provide the advantages of screening compounds with different off-rates. However, due to intricate handling techniques and instrumentation requirements, these methods have not been widely implemented. Additionally, when compared to traditional formats, the screening capacity of CE and LC is significantly lower due to limitations in injection volumes and these methods need further optimization for adaptation to large scale DEL selections.

For targets that are difficult to prepare in a purified state, cell-based selections have been developed. This strategy maintains the target in its physiological condition, with the presence of cofactors and other interacting cellular components. However, the existence of additional proteins, cellular activities, and the expression level of the target can complicate the selection process, resulting in an elevated background and potentially diminished signals. Cell surface target-based selection employs centrifugation to distinguish between DEL binders and non-binders, eliminating the need for an affinity tag or solid support. ^42–44  For selection against intracellular targets, DELs can be introduced into cells through conjugation with a CPP (cell penetrating peptide) ⁴⁵  or via microinjection ⁴⁶  /transfection to penetrate the cell membrane. The utilization of target proteins with a tag, ⁴⁵  a fused prey protein, ⁴⁶  or a target specific antibody ³²  can facilitate the separation and extraction of DEL binders (Figure 1.2D). Further details regarding cell-based selection will be elaborated in a subsequent chapter (Chapter 3 – Live cell-based DEL selections).

In addition to the solution phase DELs used in the aforementioned methodologies, bead based DELs have been developed where each bead contains many copies of a single compound and thus commonly referred to as one bead, one compound (OBOC) DELs. OBOC DELs enable the separation process to be accomplished by probing the DEL beads with the target rather than immobilizing the target on a solid support. The selection of a DEL for binding to fluorophore-labelled RNA targets, has been accomplished with bead-based DELs with the employment of FACS (fluorescence-activated cell sorting) (Figure 1.2E). ⁴⁷  In addition, bead-based DELs can be compartmentalized into droplets using microfluidics for functional screening. ⁴⁸  The copy number of each compound on beads is much higher than the copy number in the solution phase DEL selection, and therefore this strategy augments the compound concentration and potentially increases the selection signal. Release of the compound from beads with compartmentalized droplets enables compound concentrations within typical screening concentrations (mid-micromolar), and traditional activity assays can be employed that require a binding equilibrium that is driven by the compound concentration (rather than the protein concentration). The selection process, however, is prolonged due to sorting and the throughput is significantly limited by sorting speed. This topic will be discussed in detail in another chapter (Chapter 2 – One-bead one-compound (OBOC) DELs for biochemical screens).

Some proteins are highly abundant in biological samples, allowing for the direct extraction and utilization of non-recombinant target protein in DEL selections. The absence of affinity tags can be circumvented by using cyanogen bromide (CNBr) or carboxylic acid beads, which immobilize target proteins by covalent attachment to amine groups on the protein. This immobilization method has been successful in identifying trypsin and albumin binders from DELs. ⁴⁹  Alternative strategies, such as the chemical biotinylation of natural proteins, can also be employed, with subsequent immobilization achievable via streptavidin beads. Antibody or other interaction-based beads containing an affinity capture reagent to the target protein can also be utilized if available (Figure 1.3A).

Recombinant proteins are more frequently used for DEL selections than non-recombinant proteins, enabling the production of a construct with an appropriate affinity tag for use in selections and various downstream assays. The use of engineered, recombinant proteins allows for selections to be done against different domains of a target protein, the introduction of specific mutations to explore the binding pocket of interest or binding to a wild-type versus mutant form of a protein, and the confirmation of the mechanism of action (MOA) of hit compounds. Prokaryotic and eukaryotic expression systems can provide different post-translational modifications (PTMs) of targets, further aiding in the investigation of the impact of PTMs on protein status and DEL selection outcomes (Figure 1.3B). The ability to design and produce a wide range of proteins that span multiple target classes and represent different functional and/or structural states of potential drug targets has facilitated the successful application of DEL technology to many drug discovery programs.

Enzymes are generally easy DEL targets owing to their catalytic nature, the ligandability of binding pockets, and in many cases a detailed understanding of structure and function. Various studies have been published utilizing DELs for enzyme ligand discovery. ^3,50,51  While high throughput screening typically provides hits on the basis of enzymatic activity inhibition alone, DEL binding assays present unique advantages for identifying hits with a specific MOA, particularly for enzymes with multiple substrates. For enzymes with a compulsory ordered reaction mechanism, the binding of the first substrate alters the protein conformation, enabling the recognition of the second substrate. Many enzymes also catalyse reactions via a ping-pong mechanism, in which the second substrate can only bind after product release of the first substrate. ⁵²  Therefore, inclusion of different substrates can reveal specific binding pockets and enable the investigation of the enzyme in different conformational and functional states. In a specific case study, ⁵³  the target enzyme, Na-terminal acetyltransferase (Naa50), transfers an acetyl group from acetyl-CoA to acetylate the N-terminus of substrate proteins. Compared to the apo enzyme, inclusion of acetyl-CoA afforded additional signals from the selection. The hits were validated to have a specific mechanism that showed attenuated binding in the absence of acetyl-CoA. The DEL-derived ligands demonstrated potent inhibition of enzymatic activity and good engagement of intracellular Naa50. The binding mode was also explored in co-crystallization studies. The convenient inclusion of multiple DEL selection groups representing different functional and conformational states of the protein provides the opportunity for rapid investigation of a target protein at intermediate or unique conformations, thereby facilitating the identification of ligands with distinct MOAs.

In other case studies of kinases in DEL selections, Mer constructs varying in size and phosphorylation status, with and without ATP pre-incubation, along with three off-target proteins, were used in the DEL selection experiment. The variability in constructs and the presence of ATP facilitated the discernment of potential binding mechanisms. Selective inhibitors that displayed ATP-competitive and non-competitive features were identified in this study. ⁵⁴  Additionally, the non-conserved cysteine located in or proximal to the active sites provides an effective strategy for kinase inhibitor development by offering enhanced potency or additional selectivity. In the case of Janus kinase 3 (JAK3), the non-conserved Cys909 yielded the selective covalent inhibitor ritlecitinib, making it an ideal target for covalent ligands. ⁵⁵  Covalent-based DEL selection has been published on this target, ⁵⁶  as well as for Bruton’s tyrosine kinase (BTK) ⁵⁷  and peptidyl-prolyl cis/trans isomerase NIMA-interacting-1 (Pin1). ⁵⁶  The details of covalent DELs will be reviewed in another chapter (Chapter 6 – DELs with covalent warheads).

PPI targets are perceived as challenging, primarily due to the inherent nature of the interface of the protein pair. The PPI interface is both large and flat and presents unique challenges in targeting with small molecules. Peptoid libraries, ⁵⁸  which effectively mimic the interaction of proteins, and the expansive chemical space provided by small molecule DELs both augment the probability of identifying small molecules directly. As noted in this study, ⁵⁹  compounds with submicromolar potency in inhibiting lymphocyte function-associated antigen 1 (LFA-1) and intercellular adhesion molecule 1 (ICAM-1) were identified from selection of a 4.1 billion compound DEL against LFA-1.

Interleukin-2 (IL2), a cytokine that stimulates both regulatory T cells and effector lymphocytes, plays an important role in cancer immunotherapy and autoimmune diseases. IL2 interacts with its receptor IL2R, an interaction characterized by globular proteins, which has been proven challenging to inhibit. ⁶⁰  Ligands of IL2 have been identified from different selections, ^61,62  where the initial selection involving 30 000 compounds revealed a ligand with 2.5 µM K _d (dissociation constant), while the subsequent selection with a 669 240 member library yielded a ligand with 0.34 µM K _d, with the binding site located at the CD25 interface.

In another report, ⁵⁸  an “anchor motif” strategy was utilized. Tryptophan was chosen as the starting point for design and synthesis of a peptoid DEL focused on tryptophan-mimetic indole side chains due to the presence of a key tryptophan residue in the “hot spot” of PPI. Using this focused library for the discovery of PPI inhibitors, TEAD4 binders were identified, exhibiting the ability to disrupt the hTEAD4–YAP interaction, and cellular activity was validated by reducing gene expression under control of the TEAD–YAP transcription factor complex.

GPCRs belong to a broad category of drug targets. GPCRs feature 7 transmembrane domains and respond to extracellular signals by triggering intracellular signalling pathways. Therefore, GPCRs are involved in a wide range of biological processes and are one of the most common classes of drug targets. Traditional GPCR screening has been primarily focused on cell based functional screening, structure based virtual screening and protein-based screening. ⁶³  The initial application of DELs on GPCRs was hindered due to the challenges in producing high levels of recombinant GPCR protein in the appropriate conformation. However, cell-based selections were subsequently developed for directly screening GPCR ligands using cells instead of purified proteins. In this study, ⁴²  the tachykinin receptor neurokinin-3 (NK3R) was highly overexpressed in HEK293 cells and screened with 15 billion compound DELs, yielding nanomolar antagonists. DEL selection conditions were further evaluated by conjugating the positive and negative ligands with DNA tags to investigate the correlation of selection signals and target expression density. The findings indicated that a high expression level of target protein is crucial for a successful DEL screen. Nonetheless, high density is difficult to achieve as the overexpression of recombinant GPCR protein may interfere with cell health and viability.

With progression of the understanding of GPCR structure and function, protein-based DEL selections against solubilized and purified GPCRs have been increasingly reported. In this study, ⁶⁴  a protease-activated receptor 2 (PAR2) thermo-stabilized GPCR (StaR) was generated with enhanced stability in comparison to the wild-type counterpart. This protein retained the capacity to bind known PAR2 ligands. The cell-free DEL selection was performed against detergent solubilized PAR2, involving both apo protein and the pre-complexed form with the antagonist AZ6343, leading to the identification of nanomolar agonists and antagonists. In another study, purified human β2-adrenoceptor (β₂AR) was reconstituted in detergent-free, high-density lipoprotein (HDL) particles with the orthosteric site occupied by a high-affinity β-agonist, BI-167107. This strategy resulted in the identification of positive allosteric modulators (PAMs), which exhibited positive cooperativity with orthosteric agonists and high selectivity for β₂AR. ⁶⁵  In addition to PAMs, negative allosteric modulators (NAMs) have also been identified for GPCRs by DEL selections. The detergent solubilized inactive μ-opioid receptor (μOR)/naloxone and active G-protein and agonist bound receptor were both included in the selection. A NAM compound of the μOR was identified, its structural mechanism was elucidated, and it was subsequently validated with in vivo efficacy. ⁶⁶ 

In addition to conventional bead-based affinity selection and cell-based selections, the receptor-affinity chromatography (RAC) approach was developed for ligand discovery with GPCRs. Using this strategy, the GPCRs were immobilized on a solid support, packed into stainless steel columns, and DEL selection was performed against the immobilized GPCRs incorporated in the HPLC (high-performance liquid chromatography) apparatus. Ligands for β₂AR, cysteinyl-leukotriene receptor (CysLT), angiotensin II type I receptor (AT1R) and endothelin receptor A (ETAR) have been identified via this selection strategy. ^39–41 

Other strategies using GPCR-fusion protein constructs in cell-based DEL screens have been reported. The utilization of UltraID-δ opioid receptor fusion protein in Expi293F cells allowed the biotinylation of DEL molecules upon binding. Consequently, the selection could potentially improve recovery and enrichment of DEL ligands without necessitating the purification of the target protein. ³⁴  A selective, G-protein-biased agonist of the K-opioid receptor was identified using live cell DEL selection by incorporating TurboID split protein and complementation. ⁶⁷  A comprehensive review of cell-based DEL selection will be provided in another chapter (Chapter 3 – Live cell-based DEL selections).

DNA- and RNA-binding proteins primarily consist of transcription factors, DNA/RNA helicases, polymerases, nucleases, and scaffolding proteins. These proteins play an important role in regulating gene expression and maintaining genome integrity. DEL selection has been applied to various target types, but there is scepticism of using DELs for this class of proteins because of the DNA tags. The interference introduced by DEL DNA tags and DNA binding by the target potentially prevents the accurate recognition of small molecule binding, thereby hindering the effectiveness of the DEL selection. Despite the potential interference, successful applications of DEL selections to DNA- and RNA-binding proteins have been reported.

120 billion-compound DELs were screened against estrogen receptor α (ERα) WT and 3 gain-of-function (GOF) mutants. The resulting nanomolar ligand was conjugated with different E3 binder alkynes and further developed into degraders. The construct used in the DEL selection was the ligand binding domain (LBD) of ERα with the DNA binding domain deleted. Therefore, the potential DNA binding issue in the DEL selection was circumvented. ⁶⁸  Another DEL selection against the transcription factor TEAD-4 also utilized the same strategy by using a construct without the DNA binding domain. ⁵⁸ 

Monomeric RNase L has been utilized in DEL selection to identify dimer inducers and the compounds were further developed into RiboTACs (ribonuclease targeting chimeras) targeting the micorRNA-21 precursor. ⁶⁹  The construct used in the DEL selection lacked DNA processing or binding, and therefore the DNA binding effect was not mentioned.

In addition to published cases, numerous conference posters and oral presentations have highlighted ligand discovery to DNA binding proteins using DEL selections. DEL selection has been implemented on these targets at HitGen as well. The overall DNA binding effect can be alleviated or differentiated from small molecule binding by using a variety of strategies. These include the use of different constructs, DNA binding site mutations, consensus DNA blocking, inclusion of counter targets, and ligand competition for specificity checks. Control DELs containing diversity in the DNA barcodes but lacking small molecule compounds can also be useful for discriminating DNA binding from small molecule binding. Furthermore, a bioinformatics analytical algorithm is crucial for false positive differentiation of binding based on DNA sequences. ⁷⁰ 

In recent years, there has been a growing recognition of the pivotal role played by nucleic acids as druggable targets in the field of experimental therapeutics. The ability to selectively modulate the function and expression of nucleic acids has opened up new opportunities for the development of innovative therapeutic strategies. While DEL selection has been widely applied to a variety of protein targets, the utilization of DELs in nucleic acid targets remains questionable, owing to the concern that DEL DNA tags would interact with the nucleic acid target and interfere with screens designed to discover small molecule ligands. Various investigations have been applied to nucleic acid targets. The DEL selections utilized either targets labelled with biotin or poly A for immobilization and separation, or through targets labelled with a fluorophore for selection with bead-based DELs (Figure 1.4).

DEL selection against the G-quartet in the c-myc promoter has been reported using 33 distinct DELs containing 120 billion compounds. Positive binding compounds with selectivity over non-G-quartet DNA targets were identified. A parallel library with the small molecules cleaved off the DNA construct was included to help distinguish false positives due to DNA hybridization. Other strategies were not elaborated in this paper, which might be attributed to the established understanding that G-quartets possess a highly compact and stable structure. ⁷¹  This is the first report demonstrating DEL selection on a nucleic acid target, thereby encouraging more research for this application of DEL technology.

With the emergence of RNAs as therapeutic targets, DEL selection on RNA targets has been explored more intensively. A published study reported the use of a bead-based DEL to screen a library of RNA structures. RNA structures were labelled with fluorophores and sorted by flow cytometry. A base paired counter screen with a different fluorophore label was incorporated to increase selectivity and minimize false positives (two-color FACS). The paper did not provide specific details on any DNA binding interference, which may be attributed to the use of a solid-phase library in the selection process. In contrast to the solution phase DEL, the stoichiometry of DEL DNA tags and small molecules in bead-based libraries is typically not 1 : 1, with the small molecule in excess of the DNA. The utilization of an RNA library with a randomized 3 × 3 region instead of a specific RNA target helped alleviate DNA binding interference. The screen identified a ligand that was active in cellular assays for a specific RNA structure present in primary microRNA-27a. ⁴⁷  This strategy using a specific RNA target has been reported in another study where the RNA repeat expansion r(CUG)exp was used in the two-color FACS screening. Hits identified were further augmented when attached to an RNA degrading molecule. ⁷² 

Solution phase DEL presents more obstacles in DEL selection against RNA targets. Given the 1 : 1 stoichiometry of DNA tags and small molecules in most DELs, the potential effect of DNA tags on RNA structure and challenges associated with DNA binding have been difficult to tackle. This is especially concerning for less compact and stable RNA targets, which usually adopt dynamic conformations under different conditions. A comprehensive exploration of solution phase DEL selection on RNA targets has been reported. ⁷⁰  The utilization of RNA patches, which are short fragments of the RNA target and serve as surrogates to hybridize with DEL DNA tags, in combination with competitive elution, significantly reduce DEL DNA tag binding and increase the specificity of the DEL selection. In addition, informatics filtering tools such as K-mer analysis and motif search were developed to differentiate DNA binding computationally. This approach has led to the identification of specific ligands for FMN riboswitch RNA from the selection. ⁷⁰  A separate instance of DEL selection against an RNA aptamer target has been noted, but the specific details of the selection process were not disclosed. ⁷³ 

The intrinsic nature of DEL design and screening, whereby libraries of small molecules are tethered to a second moiety via a linker and screened for target binding, makes DELs an ideal technology for PROTAC (proteolysis-targeting chimera) applications. This can be achieved by separate screening on E3 ligases or POIs (proteins of interest) for ligand identification and further development of PROTAC via conjugation. Moreover, the direct discovery of PROTACs in ternary complex formation based on the selection format has been reported. ^74–76  This application is discussed in depth in another chapter (Chapter 7 – Application of DELs for E3 ligase ligand and PROTAC development). Given the large chemical space, high sensitivity and fast investigation of structure–activity relationships (SAR), DELs can also be employed in the discovery of molecular glues. A compound with a molecular glue mechanism of action may require a precise and nuanced interaction with a target protein pair. The DNA tag and linker will potentially contribute to binding of DEL compounds to the target protein pair, a possibility that needs to be meticulously evaluated in the characterization of compounds derived from molecular glue-based DEL screens. In addition, the conjugation nature of DELs and the expansive chemical space render them a beneficial tool for antibody–drug conjugate (ADC) development. DEL selection can be utilized for the identification of payloads for specific targets. The conjugation of DEL-derived molecules to an antibody has been reported; in this particular case, the tumour antigen-specific DEL binders were developed as a switch for CAR T (chimeric antigen receptor T) cell cytotoxicity and specificity. ⁷⁷ 

Despite the success in ligand discovery for numerous target classes, there is still failure in getting hits for certain target types. Intrinsically disordered proteins (IDPs) lack a fixed or an ordered three-dimensional structure under physiological conditions, and they adopt different conformations with dynamic changes. The high flexibility and the lack of well-defined binding pockets make ligand discovery on these proteins challenging. To our knowledge no successful IDP DEL selection case study has been published. Developing DELs covering a larger and more diverse chemical space, or new selection techniques or strategies may be necessary for successful application of DELs to IDPs. Transporters and ion channels are multi-transmembrane proteins involved in various biological processes and constitute a significant portion of drug targets. The lack of DEL selection reports on these target types may be due to the difficulty in preparing the recombinant proteins as well as challenges in achieving good selectivity over closely related family members. Considering these challenges, more efforts are required to further investigate the utilization of DEL selection on these target classes. This would entail not only overcoming the technical difficulties associated with the protein preparation but also developing innovative strategies to enhance the selectivity and efficacy of DEL selections.

In addition to advances in DEL chemistries and refinement of selection strategies, we anticipate that the further development of NGS technologies will significantly impact the field. The progression of NGS is expected to reduce the cost and substantially enhance sequencing capacity. This improvement may bring about a paradigm shift in DEL selection practices, by potentially changing the selection input and the number of selection rounds significantly. For example, strategies to determine compound affinity from selection data ^14,15  will be facilitated by progression of NGS technology. With current practice, binding affinity can be inferred from DNA sequence counts of enriched compounds, but for a number of reasons a strict correlation of DNA sequence counts with compound affinity does not exist. The ability to precisely determine compound potency from primary DEL selection data would represent a significant improvement of the technology. Investigation on data interpretation has been reported, ^16–19  but with the larger sequencing datasets generated with the high NGS capacity in the future, the analytical algorithms employed to process and interpret the DEL selection results will also require enhancements in both efficiency and accuracy.

Machine learning (ML) approaches have been increasingly employed in interpreting and denoising DEL selection data. ⁷⁸  Furthermore, they have significantly contributed to the discovery of ligands for a variety of targets. ^79,80  Continued development and refinement of ML models, combined with the large quantity of data provided by DEL selection, are expected to considerably reduce the associated cost and optimize the efficiency of the DEL process. In turn, this should enhance the success rate and overall impact of the DEL platform for hit discovery.