Split-and-pool synthesis does one thing no parallel synthesis does: it makes the bead, rather than the vessel, the unit of identity. Furka and colleagues described the procedure as ordinary solid-phase synthesis with a single modification — before every coupling cycle the resin is divided into equal parts, each portion is coupled with a different amino acid, and the portions are then mixed and redivided before the next cycle — introduced specifically to give a closely equimolar mixture of peptides with predetermined sequences, and demonstrated on a mixture of 27 tetrapeptides and one of 180 pentapeptides [1].
Follow a single bead through that scheme. At each cycle it sits in exactly one vessel, so it accepts exactly one residue. After n cycles it carries one sequence in as many copies as its loading allows, while the pool carries the combinatorial product of the building blocks used. Lam and colleagues turned that geometry into a screening method: random libraries of millions of peptides produced and evaluated rapidly, with acceptor-binding ligands of high affinity identified and sequenced, on the basis of what they named a one-bead, one-peptide approach [2]. Everything downstream follows from one property — the compound on a given bead is spatially resolved but chemically anonymous until something decodes it.
What a single bead physically holds
The bead is three things at once: synthesis support, screening compartment, and the only archive of the answer. Loading determines what the third role can deliver. Hintersteiner and colleagues built an integrated screening process on the observation that the picomole amount of substance contained on one isolated hit bead is sufficient for quality control, structure determination and precise affinity determination to the target protein in solution [3]. A picomole is of order 6 × 1011 molecules, which is why one bead can be divided across several destructive readouts, and why copy number is rarely the scarce resource in a one-bead-one-compound screen. Sampling is.
Designed diversity against bead count
A library has two sizes and they are rarely equal. Designed diversity is combinatorial arithmetic: a hexapeptide drawn from the twenty proteinogenic residues has 206 = 6.4 × 107 members. Physical diversity is however many beads went into the tube, discounted by whatever the synthesis failed to produce in usable amounts.
Two statistical questions follow, and they are distinct. The first is whether split-and-recombine delivered the designed compounds in near-equal amounts. Zhao and colleagues formalised the deviation from the ideal equimolar distribution of the final products as an overall relative error, showed it to be related to the Pearson statistic, established that this statistic still follows a chi-squared distribution even though split/recombine sampling differs from the schemes used in analysis of categorical data, and derived from that the number of beads required so that, with 99% confidence, the overall relative error stays below a pre-set tolerable limit [4].
The second question is coverage: what fraction of the designed library is physically present among the beads you screen. Working on microfluidic droplet screening of DNA-encoded one-bead-one-compound libraries, MacConnell and Paegel reported that Poisson statistics describe nearly all aspects of such screens, and expressed aliquot size as a library equivalent, ε — elements sampled divided by unique library members — with coverage defined as the fraction of the library observed at least k times [5]. Their worked case is blunt. At ε = 2, 86.5% of the library is present k ≥ 1 time, 59.4% is present k ≥ 2 times, and 32.3% k ≥ 3 times [5]. Two library equivalents of the hexapeptide above is 1.3 × 108 beads to handle, and roughly one designed sequence in seven is still absent from the tube. A stated library size is a property of the design; what you screened is a draw from it.
The design consequence is that positions are expensive. Adding one variable position to a library built from twenty building blocks multiplies designed diversity by twenty and, at fixed ε, multiplies the bead count you must handle by the same factor. Libraries are therefore shortened, or the building-block set is restricted, or the coverage shortfall is accepted and stated because the screen is meant to sample sequence space rather than enumerate it. What is not defensible is reporting designed diversity as though it were the number of compounds tested.
On-bead screening, and where the false positives come from
The standard format incubates the library with a labelled target and picks the beads that light up. Several distinct mechanisms put beads in that pile for reasons unrelated to the affinity you are trying to measure.
The first is avidity created by the bead surface itself. Chen and colleagues identified high ligand loading as a major source of nonspecific binding, because it permits a macromolecular target to interact with multiple ligands on the bead surface simultaneously. They built a phosphotyrosyl peptide library on spatially segregated TentaGel microbeads carrying a tenfold reduced peptide loading at the surface but normal loading in the bead interior, and recovered a specific recognition motif for each member of a panel of 10 SH2 domains. Screened against a control library with unaltered high surface loading, six of those ten domains gave sequence biases ranging from minor perturbation in the relative abundance of different sequences to the exclusive selection of false positive sequences with no measurable affinity for the target protein [6].
The second is the label. Marani and colleagues screened a one-bead-one-peptide library with streptavidin conjugated to ATTO 590 or Texas Red and sorted beads on a COPAS flow sorter, isolating both positives arising from the peptide-streptavidin interaction and false positives produced by peptide-dye interaction. The two classes were distinguishable by appearance: false positive beads showed bright homogeneous fluorescence, whereas positive beads showed heterogeneous fluorescence with a characteristic halo, greatest at the bead surface and lowest in the core, a difference more evident with Texas Red than with ATTO 590. By MALDI-TOF, most sequences from positive beads carried the His-Pro-Gln motif, while peptides from false positive beads were rich in Leu/Ile, His, Phe and Tyr [7].
The third is simply the aggregate rate. Doran and colleagues, screening one-bead-one-compound libraries on hydrophilic resins against labelled proteins and other biomolecules, report that the method has a high false positive rate: beads scored as hits often display compounds that prove to be poor ligands once they are resynthesised and carried through validation trials [8].
There is also a systematic distortion in what an on-bead screen can rank at all. Hintersteiner and colleagues determined dissociation constants for bead-immobilised ligands interacting with soluble proteins on TentaGel by confocal nanoscanning, and found that the affinity ranges sampled in on-bead screening are narrower than those of equivalent interactions in homogeneous solution; they note that on-bead screens more often than not reveal apparent high-affinity binders through strong protein complexation on the matrix of the solid support, and that after decoding, resynthesis and solution testing the primary hits can turn out to be unexpectedly weak binders or may fall outside the detection limit of the solution assay [9]. A bright bead therefore licenses one conclusion — this bead retained labelled target under these wash conditions — and not the conclusion that its compound is the tightest binder present.
Reading the hit
If the compound is a linear peptide with a free N-terminus, the bead can be sequenced directly, which was part of the original one-bead, one-peptide proposal [2]. Edman chemistry consumes the material, requires the peptide to be linear and unblocked, and returns one residue per cycle, so anything cyclic, N-capped or non-peptidic is out of its reach.
Mass spectrometry avoids the chemistry constraint by building the answer into the bead during synthesis. Youngquist and colleagues used a termination synthesis approach that encodes each resin bead with the information needed to establish the sequence of the full-length product it also carries, and read the sequences by matrix-assisted laser desorption ionisation. The same spectra expose deletion peptides, side-reaction products and incomplete-deprotection products, so the readout doubles as a direct assessment of the quality of the synthetic library. Screening an anti-gp120 monoclonal antibody against a hexapeptide library, they isolated eight active peptides, six of which possessed the exact recognition sequence for the antibody [10].
Partial Edman degradation combined with mass spectrometry extends the idea to compounds that cannot be sequenced at all. Joo and colleagues spatially segregated 90 µm TentaGel beads into outer and inner layers, displayed cyclic peptides on the bead surface and placed the corresponding linear peptide in the inner core as an encoding sequence, then determined hit identity by sequencing the interior encoding peptide by partial Edman degradation and mass spectrometry after screening. On-bead screening of an octapeptide library with a theoretical diversity of 160,000 identified cyclic peptides that bind streptavidin [11].
Encoding, and what a tag actually certifies
Encoding separates the molecule being screened from the molecule being read. Ohlmeyer and colleagues synthesised a 117,649-member peptide library on beads indexed with inert chemical tags used as a binary code recording the reaction history of each bead, read the code directly from a single bead by electron capture capillary gas chromatography, and demonstrated correct selection of library members on the basis of binding to a monoclonal antibody [12]. Brenner and Lerner proposed the nucleotide version, in which alternating parallel combinatorial synthesis encodes each member of a large chemical library with a unique nucleotide sequence that can be amplified by replication, used for enrichment of the bound molecules by serial hybridisation, and decoded by sequencing the tag [13]. Liu and colleagues kept the tag peptidic: topologically segregated bifunctional beads made by a biphasic solvent strategy carry the test molecules in the outer layer and coding tags in the bead interior where they do not interfere with screening, the coding peptide is built from unnatural α-amino acids derived from the same building blocks as the test compound so that extra synthetic steps are eliminated, and positive beads are sequenced directly by Edman degradation with no cleavage or retrieval of the tag. They generated and screened a 158,400-member encoded peptidomimetic library for ligands binding streptavidin [14].
A tag certifies synthetic history, not molecular identity. It records which vessel a bead visited at each step, which is the correct structure only if every coupling went to completion and no side reaction intervened — a failure mode the ladder-based mass spectrometric readout makes visible and a tag read cannot.
Positional scanning, and what it trades away
The alternative design abandons spatial resolution and works with soluble mixtures. Houghten and colleagues generated synthetic peptide combinatorial libraries composed of mixtures of free peptides in quantities usable directly in existing assay systems, and combined screening with an iterative selection and synthesis process; starting from a library of more than 34 million hexapeptides, they identified the antigenic determinant recognised by a monoclonal antibody [15]. Pinilla and colleagues restructured that into positional scanning: six individual positional libraries of hexamers, each with a single position defined and five positions as mixtures, screened in parallel, with the antigenic determinants recognised by two different monoclonal antibodies correctly identified upon a single screening [16].
What positional scanning buys is one round of synthesis in place of several sequential ones. What it gives up is per-molecule resolution: each defined sublibrary reports the pooled response of every sequence sharing that residue at that position, so the candidate is assembled by taking the best residue at each position independently and then making and testing that single molecule. The assumption behind that assembly has been examined experimentally. Wilson-Lingardo and colleagues compared pooling strategies on a library of 810 compounds in a phospholipase A2 inhibition assay and found that, with one exception, all iterative deconvolution strategies and position scanning resulted in identification of the same compound [17]. That is a reassuring result from a small non-peptide library with a well-behaved enzyme assay, and it is not a licence to skip the confirmation step in a peptide library many orders of magnitude larger.
A hit is not a binder
Everything above yields a candidate structure, not a measured affinity. The bead was scored under one set of wash conditions, with a multivalent surface presentation and frequently a fluorophore in the mixture; the compound has been read, not characterised. Two things move a hit toward a validated binder.
The first is replication within the screen itself. Doran and colleagues showed that deliberately redundant libraries — in which more than one bead displaying the same compound is present — largely eliminate the problem, because compounds isolated more than once are likely to be high quality ligands for the target whereas compounds isolated only once have a much higher likelihood of being poor ligands; the cost is that redundancy limits the number of unique compounds screened at one time [8]. That is the same k-class argument as the coverage statistics above, run in the opposite direction: diversity deliberately surrendered in exchange for confidence per hit.
The second is resynthesis and retest in solution, and it is not optional. The compound is made again independently, characterised, and its affinity measured in a format that does not involve the bead, because the surface presentation was part of what generated the signal — which is what the ligand-density and affinity-range results above establish. Where full-scale resynthesis is prohibitive as a triage step, the picomole of material on a single hit bead can carry those same measurements itself [3], separating the candidates worth resynthesising from those that were artefacts of the bead.
Where this sits relative to display technologies
Phage display and mRNA display are biological selections: the library is encoded genetically, sequence and molecule are linked by translation, and rounds of binding and amplification enrich the pool before anything is sequenced. Lam and colleagues framed the one-bead, one-peptide method explicitly against bacteriophage vectors carrying random oligodeoxynucleotide inserts, judging those techniques to have severe limitations [2]. A chemically synthesised library has no intrinsic amplification step, which is precisely the gap nucleotide tagging was proposed to close by restoring replication and hybridisation-based enrichment [13]; but it is also not restricted to what a ribosome will accept, and the same split-and-pool protocol builds cyclic peptide and peptidomimetic libraries [11, 14]. The practical division is therefore not size but chemistry and readout: a display selection gives enrichment and deep sequencing over ribosomal building blocks, while a bead library gives arbitrary chemistry, single-round screening, and the obligation to prove every hit again from scratch.
A report of a bead-library screen should state the designed diversity and the number of beads actually screened, the resin and its surface loading, the label and the blocking and wash conditions, the decoding method, whether the library carried redundancy and at what k a hit was called, and — separately from all of that — the solution affinity of the resynthesised compound. Without the last item the screen has produced a sequence, not a ligand.