Multiple Antigen Peptides and Branched Dendrimer Scaffolds

Lysine is the reason this architecture exists. It is a trifunctional amino acid, and both of its amino groups — the α-amine and the ε-amine of the side chain — will accept an acylation. Acylate both amines of a resin-bound lysine with a further lysine and the number of growing chain termini doubles; repeat and it doubles again. Tam's original construct used a core matrix of three sequential levels of lysine to present eight amino termini, anchored peptide antigens of 9–16 residues on them, and was assembled in a single solid-phase synthesis to a final M_r of about 10,000 ^[1].

The point of the geometry is that copy number comes from the scaffold rather than from a carrier protein. Posnett and colleagues synthesised a 14-residue sequence directly onto a branching lysine core, eight copies linked to the core through their C-terminal residue, and reported a molecular weight of 13,422 of which only 7% was contributed by the lysine residues of the core ^[2]. Antisera raised in mice and rabbits against that construct contained no antibodies directed at the central lysine core, and the same report noted that the approach extends to 16 or 32 arms, or to constructs bearing two different peptides ^[2]. That is the case against a carrier conjugate stated in its own terms: the need to conjugate peptides to a carrier protein for optimal immunogenicity brings a number of problems, including a possible immune response to the carrier itself ^[2].

What the branching actually buys

Two distinct effects get conflated here, and separating them is the reader's first job. Tam and Zavala compared monomeric peptides against the corresponding MAPs in a solid-phase radioimmunoassay: monomeric peptides of 12 and 17 residues bound poorly to plastic surfaces at concentrations up to 30 µg/ml and showed no immunoreactivity to specific polyclonal or monoclonal antibodies, while the MAP forms showed excellent binding capacity and immunoreactivity at 0.11 µg/ml, with octameric branching found to be the optimum ^[3]. But the authors attribute that shift to the MAP enhancing the coating capacity and the avidity of the synthetic peptide ^[3], and a plate-based assay cannot pull those two apart. More peptide adsorbed per well, and tighter multivalent engagement of the same amount of peptide, produce the same signal. If you are running that comparison, you need an independent measure of surface density before any part of the improvement can be assigned to avidity.

The topology does something copy number alone does not. Falciani and colleagues synthesised several bioactive peptides in monomeric, two-branched and tetra-branched form and incubated them in human plasma and serum, detecting unmodified and cleaved sequences by high pressure liquid chromatography and mass spectrometry. An increase in peptide copies did not increase resistance in linear multimeric sequences, whereas multimericity progressively enhanced the proteolytic stability of branched multimeric peptides ^[4]. Same copy number, different architecture, different outcome — a branched construct is not interchangeable with a tandem repeat.

Divergent construction, and why it degrades with generation

The divergent route builds the core on the resin, then elongates every arm in parallel with one stream of coupling and deprotection cycles. Darbre and Reymond describe a modular version: divergent synthesis on solid support alternating proteinogenic α-amino acids with branching diamino acids, producing peptide dendrimers with a molecular weight of 3–5 kDa ^[5]. Maillard and colleagues give the protocol form — a branching diamino acid used at every third position, split-and-mix solid-phase peptide synthesis, structural diversification restricted to four amino acids per variable position, yielding libraries of approximately 60,000 sequences ^[6]. The method, as they put it, resembles that for synthesising linear peptides ^[6] — which is the trap.

In a linear synthesis, one incomplete acylation produces one deletion peptide and the rest of the chain carries on. In a divergent dendrimer, every coupling step is delivered simultaneously to every arm, and a single incomplete acylation at any one of them yields a molecule that is defective while remaining nearly indistinguishable from the target. Take the deliberately naive model in which couplings at different sites succeed independently with the same probability p. A linear 14-mer at p = 0.995 gives 0.995¹⁴, about 93% intact chains. An octameric construct with the same 14-residue arm has 112 amide bonds to form in the arms alone, and 0.995¹¹² is about 57%. Drop to p = 0.99 and the linear case holds 87% while the octamer falls to 32%. The exponent is set by arm count multiplied by arm length, so it grows with generation number and the fraction of perfect molecules falls accordingly. Nothing in it is a measurement; it shows only how a per-coupling efficiency that is unremarkable in a linear synthesis becomes the dominant term in a branched one.

Keah and colleagues give the empirical version: the direct synthesis and rapid characterisation of MAPs has presented difficulties, partly because of the formation of incomplete or truncated peptide sequences during the synthetic procedure, which is why many researchers resorted to ligation procedures instead ^[7]. Their fix was a capping procedure introduced after each amino acid coupling step, which improved the yield of the desired construct significantly and meant that any deletion peptides which did form could be eliminated more readily during subsequent purification ^[7]. The gain there is separative rather than chemical: capping an unreacted amine turns what would have been a single-residue deletion into a permanently terminated chain, and a terminated chain sits chromatographically much further from the target than a deletion does.

Difficulty at the branch points is addressable. Kowalczyk and colleagues, working with a strongly self-associating viral ectodomain epitope, found the fully stepwise solid-phase approach to be quite practical, especially when 6-aminohexanoic acid spacer units providing increased internal flexibility were inserted at each branching point ^[8].

Convergent ligation of preformed arms

The convergent alternative purifies each arm as an ordinary linear peptide and then attaches preformed arms to a preformed core, so every building block has passed a purity gate before assembly. The chemistry has to be chemoselective enough to work on unprotected peptides. Spetzler and Tam exploited the specific reaction between a weak base and an aldehyde under acidic conditions: benzoyl hydrazine, or the 1,2-aminothiol of cysteine, placed at the N-terminus of an unprotected peptide reacts with an alkyl aldehyde on the core matrix to form a stable hydrazone linkage or a five-membered thiazolidine ring respectively, giving dendrimers with four or eight copies at 5–16 kDa, characterised unambiguously by laser-desorption mass spectrometry ^[9]. Elevated temperature plus a water-miscible organic co-solvent gave a combined rate enhancement of about 10-fold, completing the ligations in under 4 h instead of 2–3 days ^[9].

Oxime chemistry extends the accessible size. Rose and colleagues found that low pH, down to 2.1, was suitable for polyoxime formation, and examined oxime stability at alkaline pH to resolve a literature controversy: the bond was stable for up to 65 h at pH 8 and for more than 2 h at pH 9 ^[10]. Using natural rather than synthetic peptide building blocks they synthesised and characterised by mass spectrometry nine-component species with molecular masses above 48 kDa, about twice the size of homogeneous artificial proteins previously described ^[10].

The head-to-head comparison is worth citing when someone asks which route is cleaner. Rose and colleagues built an artificial protein containing four copies of a 23-residue peptide using oxime chemistry and compared it with two tetrameric MAP constructions of the same peptide, displayed either radially or linearly, made by conventional techniques; the tetra-oxime was much more homogeneous, yielding a single peak on reversed-phase HPLC and the correct mass spectrum ^[11].

That result does not generalise without conditions attached. In the study that recommended flexible spacers for stepwise synthesis, chemoselective thioether ligation in solution failed on the same epitope: its strong tendency to self-associate was a serious inconvenience for conjugation in solution, which as a result failed to produce the target MAPs with the specified number of copies ^[8]. Convergent assembly moves the problem from coupling efficiency to the solution behaviour of the arm, and an arm that aggregates will under-load the core. Kowalczyk and colleagues later published complete accounts of both routes using that 24-residue sequence, which they call a rather challenging epitope, and a far more manageable shortened 12-residue version; they also noted something that should shape how you read the older literature, namely that experimental reports on these constructs are most often lacking in chemical detail about their preparation and characterisation ^[12].

Why a branched product resists purity assignment

A reversed-phase separation resolves molecules by the composition and accessibility of their hydrophobic surface. A MAP whose arms are copies of one sequence presents eight near-identical surfaces, and a molecule missing one residue from one arm differs in a small fraction of that total surface. The chromatographic difference is correspondingly small, and the failure product tends to elute as a shoulder or under the main peak. The discriminating observation in the polyoxime comparison was precisely a single reversed-phase peak ^[11] — worth having, but a necessary condition for homogeneity rather than a sufficient one.

Intact mass is no better on its own, for arithmetic reasons. In the octameric construct above, M_r 13,422 with the core contributing 7% ^[2], a single missing residue in one of eight arms shifts the molecular weight by between roughly 0.4% (glycine, 57 Da) and 1.4% (tryptophan, 186 Da). Excluding that species requires mass resolution and dynamic range good enough to see a satellite less than one part in a hundred away from a large multiply-charged envelope. Worse, the defect is degenerate: a one-residue deletion produces the same mass shift whichever of the eight arms carried it, so even a resolved satellite tells you a residue is missing somewhere, not how the population of arms is distributed across molecules.

The workable answer is to stop measuring the intact object. Keah and colleagues digested the MAP construct enzymatically with trypsin and characterised the digest by reversed-phase high-performance liquid chromatography coupled to electrospray mass spectrometry, so that fragments cleaved at specific lysine or arginine sites within the sequence of the parent peptide could be readily determined and the kinetics of the digestion easily followed; the procedure confirms that all of the multi-dendritic arms of the purified construct have been equivalently elongated and that the construct contains the correct peptide sequence ^[7]. The trade is explicit. The assay is destructive, and it reports on the population of arms rather than on the composition of any individual molecule, so a digest consistent with eight correct arms is also consistent with a mixture whose defects average out.

The same logic runs through dendrimer library work, where the sequence of a dendrimer can be deduced uniquely from an amino-acid analysis of the solid support bead, an analysis described as more reliable, faster and far less costly than Edman sequencing ^[6]. In that setting a branched structure is assigned from composition plus synthetic history rather than by sequencing the assembled object.

What a defect population looks like under exhaustive measurement has been documented most thoroughly outside the peptide field, and those numbers are an analogy, not a transfer. Van Dongen and colleagues analysed generation 5 poly(amidoamine) dendrimer — a polyamidoamine built by a different chemistry, not a peptide dendrimer — using high performance liquid chromatography, potentiometric titration, mass spectrometry, size exclusion chromatography and nuclear magnetic resonance together. They identified, isolated and characterised trailing generations and dimer, trimer and tetramer species as the major structural defects, determined that the G5 material present in the as-received mixture contains 93 arms on average, and developed two model systems capable of generating the experimentally observed mass range and polydispersity at defect rates of 8–15% ^[13]. Five orthogonal techniques were needed to get there, and what came out was a distribution rather than a compound.

Reading multivalent binding data

Once a dendrimer is used as a binder rather than as a display scaffold, the numbers it generates need different handling. Mammen, Choi and Whitesides set the frame: polyvalent interactions are characterised by the simultaneous binding of multiple ligands on one biological entity to multiple receptors on another, they can be collectively much stronger than the corresponding monovalent interactions, and they can provide the basis for mechanisms of agonising and antagonising that are fundamentally different from those available in monovalent systems ^[14]. The strength is a property of the pairing of two multivalent partners, not of the ligand alone.

Kitov and Bundle put arithmetic under that. Their model describes multivalent binding via two microscopic binding energies corresponding to intermolecular and intramolecular interactions, whose relative contributions depend on the distribution of complexes with different numbers of occupied binding sites. A third term, which they call the avidity entropy, is a function of the degeneracy of bound states; it grows rapidly with the number of receptor sites and with ligand multivalency, it always favours binding, and it explains why multivalency can overcome the loss of conformational entropy when ligands displayed at the ends of long tethers are bound ^[15]. Critically, the microscopic parameters may be determined from the observed binding energies for a set of oligovalent ligands by nonlinear fitting with the model ^[15]. A single IC₅₀ or K_d measured on one dendrimer is a composite of all of those terms and licenses no statement about the affinity of an individual arm. Extracting a per-arm number requires a valency series and a model; a per-arm affinity quoted without one is an assumption wearing a number.

Valency also does not determine outcome by itself. Berthet and colleagues combined cyclopeptide (R) and lysine-based dendron (D) units in all possible arrangements following an iterative oxime ligation procedure, functionalised them successively with α-fucose and β-fucose to give hexadecavalent glycosylated scaffolds RD16, RR16, DR16 and DD16 alongside smaller tetra- and hexavalent scaffolds R4 and R6, and evaluated the influence of valency, architecture and anomeric configuration on binding to the αFuc-specific bacterial lectin LecB ^[16]. Competitive enzyme-linked lectin assays revealed that only the RD16 architecture displaying αFuc — compound 9A — reached strong binding improvement over the monovalent reference αMeFuc, with an IC₅₀ of 0.6 nM. A dissociation constant of 28 nM was measured for that same compound by isothermal titration microcalorimetry, reported as the highest affinity ligand then described for LecB, and calorimetry together with molecular modelling suggested the high affinity observed might be due to an aggregative chelate binding involving four sugar head groups and two lectins ^[16].

Three readings follow from that one study. Four scaffolds of identical valency gave different results, so valency is not a sufficient design variable and the spatial arrangement of the arms is doing work. The surface-competition IC₅₀ and the solution dissociation constant are different observables on the same molecule, and the gap between a value obtained by competition against a coated surface and one obtained by calorimetry in free solution is not something to normalise away. And the proposed bound complex is not one ligand to one receptor — four head groups engaging two lectin molecules — so any fitted dissociation constant is conditional on that stoichiometric model. Note the analyte class too: those arms are carbohydrates on a lysine dendron and cyclopeptide, so the figures describe a glycosylated branched scaffold and do not transfer to a peptide-armed dendrimer.

Divergently built peptide dendrimers do bind and turn over substrate at modest size. Darbre and Reymond report rate accelerations of k_cat/k_uncat = 90,000 obtained when the design was changed to monomeric peptide dendrimers alternating two amino acids with the branching unit ^[5], and the combinatorial protocol from the same laboratory reports substrate binding with K_M of 10–300 µM for hydrolysis of acyloxypyrene-trisulfonates in aqueous buffer ^[6]. Both figures belong to specific designs and specific substrates; neither is a property of branching as such.

What to record

For any branched construct the notebook entry has four parts: the route, meaning divergent all-solid-phase or convergent ligation and which ligation chemistry; whether capping was applied at every coupling and whether spacers were inserted at the branch points; the analytical evidence with its resolution limit stated, including whether anything orthogonal to intact mass was run; and the assay geometry behind any binding number. The last of those decides whether an avidity figure means anything off the plate it was measured on.