Peptide Pull-Downs: Bait Design, Controls and Enrichment Ratios

A peptide pull-down presents one short sequence, stripped of the protein it came from, at a local concentration deliberately set high enough to retain weak binders, to an extract in which compartments no longer exist. That is the method's power and its failure mode in the same sentence. Interactions mediated by short linear motifs in disordered regions typically have low affinities, so peptide pull-downs require low stringency conditions, which often results in many nonspecific binders — which is why quantitative mass spectrometry, rather than a gel and a band, is what makes the experiment interpretable ^[1].

What comes back is a list of proteins with ratios attached. What determines that list is the geometry of the bait, the density it is presented at, and what the lysate happens to contain. What the ratio then licenses you to claim is considerably narrower than most figure legends suggest.

The anchor decides which binders can appear at all

Fmoc solid-phase synthesis runs from the C terminus toward the N terminus, the opposite direction to biological synthesis, so standard SPOT synthesis on cellulose yields peptides immobilised via their C terminus with a free N terminus ^[1]. Acetylating that N terminus is usually worth doing: it eliminates the positive charge, which better reflects the situation within a protein sequence where the terminus participates in an amide bond, and it makes the peptide more stable ^[1]. Synthesis fidelity is part of the same argument. Coupling yields are not 100%, so the yield of correctly synthesised full-length peptide falls as length rises, which is why baits should not run much beyond about 15 residues; a capping step blocks defective chains in later cycles, trading full-length yield for truncated products and reducing the risk of false positives arising from peptides with erroneous sequences ^[1].

Anchoring direction is not a formatting detail. Recognition of C-terminal residues is one of the most common binding features of PDZ domains, and most solid-support-bound peptide libraries lack a free C terminus precisely because of C-terminal fixation to the support ^[2]. Boisguerin and colleagues developed chemistry to generate cellulose membrane-bound peptides with authentic C termini, screened a human library of 6,223 C termini against the syntrophin PDZ domain, and used the same library to identify new ligands for the ERBIN PDZ domain ^[2]. Present those sequences C-terminally anchored instead and the ligands that depend on a free C terminus are not weakly enriched; they are missing, and nothing in the downstream statistics will tell you so.

Direction is not always the dominant variable, though. Andresen and colleagues coupled biotinylated peptides to NeutrAvidin in solution, microdispensed the complexes onto activated glass, and varied the type and length of the spacer used for that presentation. Probing linear epitope peptides with their cognate monoclonal antibodies, they found that incorporating a minimum spacer length was imperative for antibody binding, whereas the peptide immobilisation direction had only secondary importance for antibody affinity and binding ^[3]. The two findings address different failure modes: direction governs whether a terminus-reading domain can engage the bait in principle, while spacer length governs whether a site close to the anchor is sterically reachable at all. Both are properties of your construct rather than of the motif, and neither is visible in the output.

Bait density is a dial you set, and it moves the answer

High immobilisation density is the reason the format works. A SPOT array spot is roughly 2–3 mm across with a typical yield of 5–10 nmol, which is 8–17 µg for an average 15-mer, and that high local concentration is what allows proteins with low binding affinities to be enriched ^[1]. Beads coupled with a synthetic peptide present the same argument in a different geometry.

The crowding that recovers weak binders also manufactures avidity. Gori and colleagues used a chemoselective click reaction to immobilise epitope sequences in linear, branched and tandem forms on a microarray and on an SPR chip carrying the same clickable coating, and reported that fine tuning of the surface peptide density plays a crucial role in exploiting oriented and multiple peptide display; by surface plasmon resonance imaging the tandem construct gave an approximately twofold increased binding capacity relative to the linear and branched peptides, which the authors attributed to a lower dissociation rate ^[4]. That was an antibody analyte on a planar surface rather than a lysate on beads, so the numbers do not transfer, but the mechanism does: a prey carrying two binding sites, or arriving as part of an assembly, can engage more than one bait when baits sit close together, and what it forms then comes apart more slowly than the monovalent interaction would. Bait density on a bead is the same dial as ligand density on a biosensor chip, and it should be recorded in the methods for the same reason.

What the lysate holds, and what the buffer costs you

The extract defines the answer set before any binding happens. Eberl and colleagues built a label-free histone peptide pull-down workflow and applied it to extracts of mouse brain, liver, kidney and testis, identifying known and putative readers of H3K4me3 and H3K9me3, most of which were ubiquitously expressed, alongside specialised readers in tissues such as testis ^[5]. A binder absent from the extract is a false negative no statistical model can recover, and a binder present only in one tissue will look sequence-independent if that is the only extract you run.

Concentration and temperature follow from the low affinities. A rather concentrated extract, on the order of 5 mg protein per ml, increases sensitivity; pull-downs are generally run at 4 °C with protease inhibitors, and the final washing buffer should omit protease inhibitors so that the subsequent digestion is efficient ^[1]. If the bait contains cysteine, disulfide formation between peptide and protein can generate false positives, which argues for a reducing agent during incubation — provided it is compatible with the chemistry holding the peptide to the support ^[1].

Wash stringency is the central trade, and the affinity-enrichment answer is to stop fighting it. Keilhauer and colleagues make no attempt to purify complexes to homogeneity, running single-step enrichment in which each pull-down contains around 2,000 background binders, and then use that background three ways: for accurate normalisation, as the comparison set so that interactors are called against other tagged samples rather than one untagged control, and as intensity profiles across all samples for validating candidates ^[6]. Those experiments used GFP-tagged proteins expressed endogenously in budding yeast, so the count is not a target for your system; the design principle — that background is information rather than dirt — is what carries over to a peptide bait, where gentle washing is not optional.

Why a beads-only control does not settle anything

Begin with what the matrix alone does. Trinkle-Mulcahy and colleagues pooled 27 independent SILAC pull-down experiments on 11 separate GFP fusion proteins in whole-cell, cytoplasmic and nuclear extracts and defined the proteins binding to the affinity matrix as a bead proteome; for Sepharose that set comprises histones, hnRNP proteins, heat shock proteins, ribosomal proteins, translation and initiation factors, DEAD box proteins and multiple cytoskeletal proteins, with over 100 further proteins outside those classes ^[7]. Comparing Sepharose, agarose and magnetic beads by three-state SILAC, they found magnetic beads gave lower backgrounds than Sepharose in nuclear extracts while Sepharose beat magnetic beads in cytoplasmic extracts, where agarose was lower still, and concluded that no single type of bead is ideally suited to all applications ^[7].

Aggregation helps where one experiment cannot. The CRAPome was built on the observation that negative controls are largely bait independent, so pooling them across many affinity purification studies increases coverage and improves the characterisation of the background associated with a given protocol; the authors note that many small-scale studies do not capture a complete, accurate background protein set when available controls are limited ^[8]. A repository tells you when a hit is something that shows up in everyone's data. It does not tell you whether the protein bound your sequence.

That is the limit of a beads-only channel. Subtracting it removes the matrix contribution and leaves intact every protein that adhered to the bait's own charge, hydrophobicity and free terminus. The controls that address the bait are sequence controls. The original design pairs an active bait with a control differing only in the feature under test: Schulze and Mann used synthetic peptides in phosphorylated and non-phosphorylated forms and read specificity directly from SILAC ratios, finding that a tyrosine-phosphorylated peptide of the epidermal growth factor receptor specifically retrieved Grb2, and that a proline-rich sequence of Son of Sevenless also specifically bound Grb2 and retrieved only SH3-domain-containing proteins as specific binding partners — demonstrating that the screen maintains specificity with low affinity interactions ^[9].

Lange and colleagues ran the same logic on beads, coupling phosphotyrosine motifs covering Y595, Y625 and Y771 of the adapter protein ADAP, together with the corresponding non-phosphorylated sequences, covalently to agarose and incubating them with Jurkat T cell lysates; comparing SILAC against enzymatic ¹⁸O labelling, they report both techniques well suited for distinguishing phosphorylation-specific peptide-protein interactions from the background ^[10].

A scrambled or mutated bait tests a different proposition — whether residue order matters or only composition — and mutated or unmodified peptides are the suitable negative controls for parallel control pull-downs ^[1]. Klingberg and colleagues pushed the idea to its stereochemical limit, running differential pull-downs under a SILAC scheme with histone H3 tails synthesised from L- and from D-amino acids; the comparison identified HAT1 and RBBP7 as part of the network interacting with the unmodified H3 tail, while a bait carrying the non-hydrolysable phosphoserine mimic phosphonomethylene-alanine at S10 recruited several isoforms of the 14-3-3 family and blocked recruitment of HAT1 and RBBP7 ^[11]. A D-peptide preserves composition, length and gross physicochemistry and alters only the three-dimensional arrangement of the side chains, which makes it about as clean a negative as a stereospecific recognition event admits.

Reading the quantitative output

Ratios run in both directions, and a workflow that only reports enrichment discards half the measurement. The tissue screen described above also found that a large class of proteins was specifically repelled by H3K4me3, and reached near-saturation of direct interactors ^[5]. Depletion relative to the control bait is a result about the modification, not a failed pull-down.

Scaling the bait side changes what the quantitation has to support. Dittmar and colleagues spot-synthesised linear tiling peptides across C/EBPβ, with and without post-translational modifications, incubated the arrays with nuclear extracts and identified interaction footprints of about 1,300 proteins ^[12]. A tiled series carries its own internal comparison, since the shape of the enrichment profile along the sequence is the evidence; a protein enriched uniformly across every window is reporting on the support, not on any one motif. That is also the point where the peptide pull-down and the peptide microarray converge as formats, and where the array's substitution logic starts to do the work.

On-bead or eluted digestion

Digesting on the beads avoids an elution step and its losses, but the immobilised ligand goes into the digest as well. Hollenstein and colleagues preincubate beads with Sulfo-NHS-acetate, which acetylates lysine residues and makes bead-bound ligands such as antibodies or streptavidin insusceptible to Lys-C-mediated proteolysis, then digest in two steps: Lys-C on-bead, followed by in-solution trypsin digestion of the released proteins. The protocol strongly reduces contaminating ligand peptides, which allows significantly higher amounts of sample to be subjected to LC-MS and improves sensitivity and quantitative accuracy ^[13].

A peptide bait presents a sharper version of the same problem. Peptides on cellulose are typically not eluted and so are not identified in the mass spectrometer, but a bait containing a cleavage site for the protease in use — arginine or lysine for trypsin — will itself be digested and become an abundant peptide species in the measurement, which is why those sequences should be included in the protein sequence database the spectra are searched against, to avoid misidentifications ^[1]. The asymmetry matters: an antibody ligand's codigestion products are a suppression problem, whereas bait fragments are also a search-space problem, because they map onto the very protein the motif was taken from.

What an enrichment ratio licenses you to say

A ratio is the amount recovered on the bait divided by the amount recovered on the control, at one bait density, one extract concentration, one incubation time and one wash. It confounds abundance in the extract with association and dissociation rates and with whatever avidity the surface supplied. It is not a dissociation constant, and replication does not convert it into one.

Getting an affinity requires a series rather than a single purification. Sharma and colleagues determined dissociation constants for cellular target proteins interacting with an externally introduced ligand by combining quantitative mass spectrometry with a defined set of affinity purification experiments, demonstrating the approach with a tyrosine-phosphorylated peptide among the affinity ligands used ^[14]. The constant comes from how recovery changes across the set; no member of that set yields it alone.

The array literature arrived at the same ceiling from the other side. Weiser and colleagues systematically compared dissociation constants of antibody-peptide complexes with SPOT signal intensities using a monoclonal antibody as the model probe and a set of peptides of differing affinity. They concluded that the technology is accurate for assigning measured spot intensities to three binding-affinity classes, that the dissociation constants of the intermediate region fall between pK_dis 5 and 7, that altering the experimental parameters causes a directed change of that region, and that the applied concentrations of the binding partners do not directly correspond to the effective concentrations in the assay ^[15]. Coarse ranking, tied to your own conditions, is what a surface-binding signal supports.

Two further limits are worth stating explicitly on any hit list. A peptide result is a claim about a motif, not about a complex: Vermeulen and colleagues determined the proteins binding trimethyl-lysine histone marks by high-accuracy quantitative mass spectrometry, then assigned those readers to complexes by interaction proteomics on full-length BAC-GFP-tagged proteins — a separate experiment, on a different bait ^[16]. And the assay is in vitro throughout: combining cell lysates with the bait can bring together interaction partners that never meet under physiological conditions, so specific follow-up experiments under more physiological conditions are required before any biological conclusion is drawn ^[1].

Where a hit goes next is set by what the pull-down structurally cannot give. A biosensor measures association and dissociation rate constants on the same immobilised peptide with surface density under your control, which is precisely the quantity the ratio confounds. A peptide array measures the sequence dependence, telling you which positions the enrichment rests on. The pull-down remains the only one of the three that starts from a whole proteome, and that is why it goes first — but on its own it ranks, it does not measure.