DNA-Encoded Chemical Library

A DNA-encoded chemical library (DEL)—also termed DNA-encoded library technology (ELT)—is an ultra-high-throughput combinatorial hit-generation platform in drug discovery in which small-molecule ligands are covalently attached via a flexible chemical linker to unique double-stranded or single-stranded DNA oligonucleotide tags that serve as amplifiable barcodes recording each compound's synthetic building blocks. Conceptualized by Sydney Brenner and Richard Lerner in a 1992 theoretical paper on "encoded combinatorial chemistry" and reduced to experimental practice in the 2000s by Dario Neri (ETH Zürich), David R. Liu (Harvard University), and Praecis Pharmaceuticals (later acquired by GlaxoSmithKline), DEL technology enables researchers to synthesize and affinity-screen 10⁸ to 10¹² distinct chemical structures simultaneously inside a single microcentrifuge tube.

Whereas conventional high-throughput screening (HTS) tests individual compounds in separate microplate wells—requiring robotic infrastructure, milligrams of each library member, and tens of milligrams of protein target to screen ~10⁶ compounds—a pooled DEL affinity selection consumes only 5 to 50 μg of target protein in a 50–100 μL volume while sampling chemical space orders of magnitude larger.

Combinatorial Scaling via Split-and-Pool Encoding

The vast numerical diversity of a DEL arises from iterative split-and-pool ("split-and-mix") synthesis coupled to enzymatic DNA ligation:

1. Cycle 1 (Split, Tag, React, Pool): An initial starting oligonucleotide ("headpiece" containing a primer-binding site and a free amine or carboxylic acid handle) is divided equally across N₁ wells (for example, N₁ = 1,000 wells of 96- or 384-well plates). In each well i, T4 DNA ligase attaches a unique short DNA codon (C_1,*i*, typically 8–12 base pairs with error-correcting Hamming distance ≥ 3), and a corresponding chemical building block (BB_1,*i*) is coupled to the chemical linker. All 1,000 wells are then pooled into a single vessel and purified by ethanol precipitation or reverse-phase HPLC.
2. Cycles 2 and 3: The pooled mixture is split again across N₂ = 1,000 wells, encoded with second-cycle DNA codons (C_2,*j*), reacted with building blocks (BB_2,*j*), and re-pooled. Repeating the process for a third cycle (N₃ = 1,000) yields a combinatorial product of:

N_total = N₁ × N₂ × N₃ = 10³ × 10³ × 10³ = 10⁹ (1 billion) barcoded compounds obtained from only 1,000 + 1,000 + 1,000 = 3,000 individual chemical and enzymatic reactions. Keeping library construction to 2 or 3 synthetic cycles prevents "molecular obesity" and ensures the resulting small molecules remain close to Lipinski's rule of five physical property space (MW 350–550 Da).

DNA-Compatible Aqueous Organic Reactions and Architecture Variants

Because the covalently attached DNA barcode is present throughout synthesis, every chemical reaction must proceed in aqueous-organic cosolvent mixtures (typically 20%–80% water with acetonitrile, DMF, or DMA) at mild temperatures (T ≤ 60–80 °C) and moderate pH (pH 4.5–10.0). Strong Brønsted acids (pH < 3.5), strong oxidants, and harsh Lewis acids cause depurination—hydrolytic cleavage of the N-glycosidic bond of adenine and guanine bases—followed by β-elimination of the phosphodiester backbone, which destroys the amplifiability of the DNA template. Over the past decade, medicinal chemists have expanded the on-DNA reaction repertoire far beyond classical amide bond formations (mediated by DMTMM or EDC/HOAt) and nucleophilic aromatic substitutions (S_NAr) to include:

• Transition-metal-catalyzed cross-couplings: aqueous Suzuki–Miyaura, Sonogashira, Heck, and Buchwald–Hartwig C–N couplings using water-soluble palladium precatalysts.
• Reductive amination, Cu(I)-catalyzed azide–alkyne cycloaddition (CuAAC click chemistry), Wittig olefination, and benzimidazole/heterocycle condensations.
• Photoredox-catalyzed decarboxylative radical couplings and Giese additions driven by visible blue LEDs at room temperature.

Two specialized architectural variants complement standard single-pharmacophore DELs:

• DNA-Templated Synthesis (DTS): Developed by David Liu, DTS hybridizes reagent-linked oligonucleotides to a template strand, raising the effective molarity of the reactants into the millimolar range so that intramolecular reactions occur selectively between sequence-programmed partners.
• Encoded Self-Assembling Chemical (ESAC) Dual-Pharmacophore Libraries: Developed by Dario Neri, ESAC pairs two complementary single-stranded DNA sub-libraries (A and B, each containing 1,000 fragments) that hybridize into a duplex displaying two fragments simultaneously to adjacent protein pockets (10³ × 10³ = 10⁶ combinations), functioning as a high-throughput fragment-based drug discovery linking engine.

Parallel Affinity Selection and Mode-of-Action Profiling

A DEL screen is executed as an affinity selection rather than a catalytic turnover assay. The pooled library (typically containing 10⁵ to 10⁶ copies of each individual barcoded molecule) is incubated in buffer (50–100 μL) with an immobilized target protein—such as a polyhistidine (His₆)-tagged protein captured on Ni-NTA magnetic beads or an AviTag-biotinylated protein captured on streptavidin beads. Stringent buffer washes remove non-binding library members (which constitute >99.99% of the pool), and thermal denaturation (heating to 75–95 °C) elutes the high-affinity binders. Repeating the selection for 2 to 3 rounds enriches genuine target ligands with equilibrium dissociation constants (K_d) in the nanomolar to low micromolar range.

Because DEL selections can be multiplexed across 8 to 96 parallel microcentrifuge tubes in a single afternoon, researchers routinely run parallel mode-of-action conditions simultaneously:

1. No-Target Matrix Control: Empty Ni-NTA or streptavidin beads alone to identify and subtract matrix-binding promiscuous compounds.
2. Apo Target Protein: Identifies binders across all accessible pockets on the macromolecule.
3. Target + Saturating Active-Site Competitor (or Cofactor): Incubating the protein with a known orthosteric inhibitor or substrate (such as ATP) blocks the catalytic cleft; compounds enriched on the apo protein but depleted in the presence of the competitor are classified as orthosteric active-site binders, whereas compounds enriched equally in both conditions bind to a distinct allosteric pocket.
4. Off-Target Anti-Targets: Parallel selection against closely related paralog proteins identifies selective scaffolds at the primary screening stage.

Next-Generation Sequencing Readout and Off-DNA Resynthesis

Eluted DNA barcodes are amplified by polymerase chain reaction (PCR) using sample-specific index primers and decoded via massive parallel sequencing (Illumina sequencing generating 10⁷ to 10⁸ sequence reads per selection condition). Each sequence read is translated back into its synthetic building-block coordinates (BB₁, BB₂, BB₃), and the enrichment factor (EF) is calculated by normalizing the observed sequence count against the pre-selection library input distribution and no-target bead controls.

Results are visualized on a three-dimensional cubic scatter plot whose x, y, and z axes correspond to the indices of Building Block 1, Building Block 2, and Building Block 3, with sphere size and color scaled to sequence read count. Genuine structure–activity relationships (SAR) appear as lines or planes inside the cube: for example, a line of enriched dots sharing the same (BB₁, BB₂) coordinates across dozens of different BB₃ building blocks reveals that the two-cycle truncated byproduct (BB₁–BB₂) is the minimal pharmacophore driving affinity. Top-enriched structures are then prioritized for off-DNA resynthesis (replacing the DNA attachment point with a small methylamide or alcohol cap) and validated in orthogonal biochemical and SPR/ITC assays. Landmark clinical candidates discovered directly from DEL screens include the soluble epoxide hydrolase (sEH) inhibitor GSK2256294, the receptor-interacting protein kinase 1 (RIPK1) benzoxazepinone inhibitor GSK2982772 (where DEL discovered a primate-specific allosteric pocket), and numerous ATX and DDR1 inhibitors.

Why can a compound show high sequence count enrichment in a DEL screen but fail to bind when resynthesized "off-DNA"?

Three mechanisms can cause discrepancies between on-DNA enrichment and off-DNA activity: (1) Synthetic byproduct active species, where an incomplete coupling or unexpected side product formed during aqueous on-DNA synthesis (rather than the nominal designed structure) is the true binder; (2) cooperative DNA–protein interactions, particularly when screening nucleic-acid-binding proteins, transcription factors, or positively charged basic surfaces (which can be suppressed by adding excess sheared salmon sperm DNA as a blocking agent during selection); and (3) linker participation, where the poly(ethylene glycol) or amide attachment linker makes favorable van der Waals contacts with the rim of the protein pocket that are lost if the linker is cleaved completely during off-DNA resynthesis.

How can DEL screens be performed in solution without immobilizing the target protein on solid beads?

Solid-phase immobilization on magnetic beads can occasionally denature delicate membrane proteins or occlude binding epitopes near the affinity tag. Solution-phase DEL platforms—such as Interaction Determination Using Unpurified Proteins (IDUP), Binder Trap Enrichment (BTE) inside water-in-oil emulsions, and cross-linking photoaffinity DELs—allow the barcoded library and target protein (even in crude cell lysates or on live cell surfaces) to equilibrate in homogeneous solution before covalently cross-linking or primer-extending only those DNA barcodes residing in immediate proximity to the target protein.

Why is library size above 10¹⁰ compounds constrained by Avogadro's number and sequencing depth?

To obtain statistically reproducible Poisson sampling during affinity selection, each individual library member must be represented by at least 10⁴ to 10⁵ molecular copies in the selection tube. For a 10¹⁰-member library at 10⁵ copies per compound, the tube must contain 10¹⁵ total DNA-tagged molecules (~1.7 nmol total DEL, or ~34 μM total DNA concentration in a 50 μL selection volume), approaching the viscosity and solubility ceiling of aqueous DNA solutions. Furthermore, decoding enrichment across >10¹⁰ compounds requires hundreds of millions of sequencing reads to distinguish true binders from stochastic Poisson noise.

1. Brenner, S., & Lerner, R. A. (1992). "Encoded Combinatorial Chemistry." Proceedings of the National Academy of Sciences, 89(12), 5381–5383. https://doi.org/10.1073/pnas.89.12.5381
2. Clark, M. A., Acharya, R. A., Arico-Muendel, C. C., et al. (2009). "Design, Synthesis and Selection of DNA-Encoded Small-Molecule Libraries." Nature Chemical Biology, 5(9), 647–654. https://doi.org/10.1038/nchembio.211
3. Goodnow, R. A., Jr., Dumelin, C. E., & Keefe, A. D. (2017). "DNA-Encoded Chemistry: Enabling the Deeper Sampling of Chemical Space." Nature Reviews Drug Discovery, 16(2), 131–147. https://doi.org/10.1038/nrd.2016.213
4. Harris, P. A., Berger, S. B., Jeong, J. U., et al. (2017). "Discovery of a First-in-Class Receptor Interacting Protein 1 (RIP1) Kinase Specific Clinical Candidate (GSK2982772) for the Treatment of Inflammatory Diseases." Journal of Medicinal Chemistry, 60(4), 1247–1261. https://doi.org/10.1021/acs.jmedchem.6b01751