Half-life extension chemistry has one job: keep an intact molecule in circulation longer without changing what it does at its target. A 2016 review of half-life extended biotherapeutics, surveying strategies already used in approved drugs and drugs then in clinical development, says the approaches it covers include those aimed at increasing the hydrodynamic radius of the biotherapeutic and those that further implement recycling by the neonatal Fc receptor (FcRn), and reports that at the time of writing half-life extension was dominated by albumin binding or fusion, fusion to an immunoglobulin Fc region, and PEGylation [1]. Fatty acid acylation, albumin fusion, Fc fusion, PASylation, XTEN and glycoengineering are all variations on one of two moves: make the molecule hydrodynamically bigger, or borrow a protein that FcRn already rescues from degradation. Several do both.

That framing is worth keeping in view, because the trade-offs follow the mechanism rather than the platform name. A strategy that works by bulk tends to pay in receptor affinity and in the fraction of the manufactured mass that is not the active part. A strategy that works by recycling pays in expression system, product heterogeneity, and the biology of whatever carrier it borrowed. Nothing here is free, and the currency differs by mechanism.

Albumin-binding fatty acid chains

Acylation attaches a fatty acid, usually through a linker to a lysine side chain, so that the peptide associates reversibly with serum albumin and circulates largely as a bound fraction. It is the most common half-life chemistry in peptide work rather than protein work for a practical reason: the product stays small enough to be built by solid-phase synthesis and characterized by ordinary small-molecule analytics.

What it costs is receptor affinity. In the report describing the design of semaglutide, increased albumin affinity and full stability against metabolic degradation were pursued together; the selected analogue had a GLP-1 receptor affinity of 0.38 ± 0.06 nM, which the authors describe as three-fold decreased compared with liraglutide, while its albumin affinity was increased, and its plasma half-life was 46.1 h in minipigs following intravenous administration [2]. That is the trade in a single sentence. The same hydrophobic group that buys residence time is competing for space the receptor wants, and the two properties were optimized against each other rather than independently.

Acylation is also not a bolt-on modification. Work on fatty diacid acylation of PYY(3-36) states directly that it is not trivial to prolong the half-life of peptides by lipidation and still maintain high potency and good formulation properties, and shows for that peptide that the position in the backbone, the type of fatty acid, and the linker each strongly influence pharmacokinetics and potency [3]. The same study makes a point that is easy to skip: understanding the proteolytic stability of the backbone is key to obtaining long half-lives by lipidation, because backbone cleavage still occurs while the peptide is associated with albumin [3]. Albumin is a depot, not a shield. A sequence with an exposed cleavage site will still be cut; what changes is how much of it is available to be cut at any moment.

How tight should the binding be?

The intuition that tighter albumin binding must give a proportionally longer half-life does not survive direct measurement well. A bispecific single-chain diabody fused to an albumin-binding domain from streptococcal protein G was compared in mice against derivatives with increased and with decreased affinity for albumin, plus a construct carrying two copies of the domain. All showed strongly increased half-life relative to the unfused diabody, and the higher-affinity variant did give a longer half-life than wild type while the lower-affinity variant gave a shorter one — but the authors describe these changes as only moderate, and conclude that in mice the half-life-extending property of that domain is only weakly influenced by affinity for serum albumin or by valency of albumin binding [4]. The two-domain construct came out almost identical to the single-domain one. That is one scaffold in one species, not a general law, but it is a reason not to treat affinity as the optimization target by default.

Albumin fusion

Genetic fusion to albumin trades a reversible association for a covalent one and recruits albumin's own FcRn-mediated recycling. The returns on engineering that interaction have been quantified. In a factor IX-albumin fusion, an engineered albumin variant conferred human FcRn binding with a dissociation constant of 0.5 nM against 164.4 nM for the wild-type fusion — about 330-fold tighter, taking the wild-type dissociation constant as the reference — and in human FcRn transgenic mice the engineered construct showed a half-life of 2.7 days versus 1 day for the wild-type fusion [5].

Two things are worth holding onto there. About 330-fold in affinity returned about 2.7-fold in half-life, which points the same way as the albumin-binding-domain result above, although the two experiments engineer different interactions — one the binding of a domain to albumin, the other the binding of albumin to FcRn. And the comparison was not single-variable: the engineered construct also carried a novel cleavable linker and the factor IX Padua substitution, which the authors report as conferring sevenfold higher specific activity [5]. Attributing the entire half-life difference to the FcRn engineering requires assuming the other two changes are neutral for clearance, which that report does not establish.

Fc fusion

Fusion to an immunoglobulin Fc is the most direct route to FcRn recycling, and it is the one case in this guide where the causal chain has been closed rather than inferred. A recombinant factor VIII-Fc fusion showed an approximately two-fold longer half-life than unfused factor VIII in hemophilia A mice and dogs; the extension was abrogated in FcRn knockout mice and restored in human FcRn transgenic mice, and the authors report that the Fc fusion had no impact on factor VIII-specific activity [6]. Knockout plus rescue is the standard a mechanism claim should be held to. Correlation between a modification and a longer half-life, with no knockout or rescue arm, is a weaker form of the same claim.

The structural cost is that Fc normally arrives as a dimer. That doubles the appended mass and can convert a monovalent binder into a bivalent one with avidity the original design never had — and the dimer is not simply packaging that can be discarded. Point mutations at the CH3–CH3 interface of IgG4 Fc abolished Fc dimerization and produced monomeric-Fc half-antibodies that were soluble, thermodynamically stable and monomeric, but with significantly reduced FcRn binding in vitro, which the authors take to suggest that avidity gains in a dimeric Fc are critical to optimal FcRn binding. The monomeric format still showed an increased terminal serum half-life compared with that expected for most alternative antibody fragments [7]. So breaking the dimer to recover monovalency costs part of the very mechanism the Fc was added for.

PASylation and XTEN

PAS and XTEN are genetically encoded, conformationally disordered polypeptides that expand hydrodynamic volume with no conjugation step at all. They were built for a specific failure mode: fast clearance from circulation via kidney filtration, which the PASylation report names as a major limitation of biopharmaceutical proteins [8]. PAS sequences are composed of proline, alanine and serine, described as hydrophilic, uncharged biological polymers with biophysical properties very similar to poly(ethylene glycol); because the fusion is genetic, fully active proteins can be produced in E. coli without in vitro coupling or modification steps, and the authors report the polypeptides to be biodegradable, stable in serum, and lacking toxicity or immunogenicity in mice [8]. XTEN is the same concept with a different residue alphabet. Its founding report used an unstructured 864-amino-acid polypeptide, showed that truncating the sequence regulates the resulting half-life, and noted that XTEN lacks the hydrophobic residues that often contribute to immunogenicity and complicate manufacture [9].

Tunability is the strongest claim these platforms make, and it has been tested comparatively. A DARPin targeting EpCAM was fused to three different sizes each of PAS and of XTEN and studied in a mouse tumor xenograft model. The authors report a clear linear relationship between hydrodynamic radius and serum half-life across both polypeptide types, reaching a half-life of up to 21 h in mice, with tumor uptake directly proportional to half-life and size; unexpectedly, charge had no influence on any parameter measured, including kidney elimination kinetics [10]. Note which variable the relationship is stated for: hydrodynamic radius, not residue count. Chain length and hydrodynamic radius are not the same axis, so a design that doubles residues is not buying double the radius.

The same 864-residue report is the source of the number most often quoted for XTEN, and it should be quoted the way its authors wrote it. Allometric scaling suggested that fusing XTEN to exenatide should increase the exenatide half-life in humans from 2.4 h to a projected 139 h [9] — about 58-fold, taking the 2.4 h figure as the denominator. That is the output of a scaling model, not a measurement, and the paper presents it as a projection.

Comparing platforms on one scaffold

Head-to-head comparisons on a single molecule are rare and worth more than arithmetic across papers. One study placed PAS polypeptides, XTEN polypeptides and an albumin-binding domain on the same anti-HER2 affibody-drug conjugate. Half-lives in mice ranged from 7.3 h for the construct including the shorter PAS to 11.6 h for the construct including the longer PAS, a spread of about 1.6-fold, with the albumin-binding-domain construct between them at 9.0 h. Yet the highest absolute tumor uptake belonged to the albumin-binding-domain construct — 60 to 160 percent higher than the PASylated or XTENylated constructs, measured against those constructs as the reference — and it also gave the best overall tumor-to-normal-organ ratios. All three extensions lowered affinity for HER2 to some extent, although the authors report that this did not negatively affect cytotoxic potency [11].

Two results in this section sit in tension. In the DARPin system, tumor uptake tracked half-life and size [10]; in the affibody system, the construct with the longest half-life was not the one with the highest uptake [11]. Different targets, different scaffolds, different payloads, different sizes — neither result refutes the other, and neither generalizes. Half-life is not a stand-in for biodistribution, and a platform ranked first on one axis can rank third on another.

Glycoengineering

Glycans act on both mechanisms at once. They add mass and hydrodynamic volume, sialic acid adds negative charge, and the terminal sugar influences whether receptor-mediated uptake pulls the molecule out of circulation: terminal GalNAc with low sialylation, the pattern typical of proteins made in HEK 293 cells, is reported to often trigger rapid clearance through binding to the asialoglycoprotein and mannose receptors [12]. Changing the terminal sugar is therefore a clearance intervention in its own right. Knocking out the GalNAc transferases B4GALNT3 and B4GALNT4 in FreeStyle 293-F cells by CRISPR/Cas9 gave a line whose factor VII-albumin fusion carried no terminal GalNAc and showed increased sialylation; that material bound less efficiently to both the asialoglycoprotein receptor and the mannose receptor in vitro, and showed improved recovery, terminal half-life and area under the curve in rat pharmacokinetic experiments. Overexpressing the sialyltransferases ST6GAL1 and ST3GAL6 in the knockout background raised sialylation further — for ST6GAL1, to the level of human plasma-derived factor VII [12].

Adding glycosylation sites is the other half of the approach, and it reproduces the potency dissociation already seen with acylation. Introducing an N-terminal extension bearing two new N-linked sites onto the alpha-chain of FSH produced a mutein glycosylated at both new sites, with increased size and charge — factors the authors note are known to reduce renal clearance of proteins — and a 3- to 4-fold increased serum half-life compared with wild-type recombinant FSH. In spite of a lower in vitro activity, that mutein showed markedly increased in vivo potency in a rat bioassay [13]. In vitro and in vivo potency moved in opposite directions in the same molecule. That is the clearest demonstration in this guide that a receptor-binding number and an exposure number answer different questions and cannot be substituted for one another.

What each strategy trades

  • Fatty acid acylation. Synthetically accessible, small, cheap to characterize; costs receptor affinity [2], is sensitive to attachment position, acyl type and linker [3], and leaves the backbone open to proteolysis even while albumin-bound [3].
  • Albumin fusion. Recruits FcRn recycling in one covalent construct; requires a eukaryotic expression system, and affinity engineering returns far less half-life than the affinity gain suggests [5].
  • Fc fusion. The best-evidenced recycling mechanism here [6], but the dimer is part of the mechanism [7], so appended mass and valency are not independent design parameters.
  • PAS and XTEN. Genetically encoded, monodisperse, tunable by length, no conjugation chemistry [8, 9]; the tuning axis is hydrodynamic radius rather than residue count [10], and a large fraction of the expressed mass is inert.
  • Glycoengineering. Acts on size, charge and receptor recognition simultaneously [12, 13]; the cost is that glycan structure is a property of the cell line and the process rather than of the sequence, so controlling it means controlling manufacture.

What is not settled

Several things here are convention rather than evidence, and saying so is more useful than a clean summary. The relationship between binding affinity and the half-life it returns looks weak in the few systems reported here, though the evidence is of two different qualities: for albumin binding it was measured across matched single-domain variants whose authors call the effect only moderate [4], while for FcRn binding it can only be inferred from a comparison that also changed the linker and the coagulation factor [5]. Those measurements are few and the animals are mice. Cross-species projection is the least tested link in the chain that motivates platform selection: the 139 h figure for an XTEN fusion is explicitly the output of allometric scaling [9], and those scaling assumptions are easy to lose when the number is quoted on its own.

Two reporting habits deserve routine suspicion, and neither has a primary source to cite because both are bench convention rather than a published finding. First, a terminal half-life is an estimate whose value depends on how long sampling continued and on where the assay's lower limit of quantitation falls; two laboratories can report different half-lives for the same molecule without either being wrong. Second, a bioanalytical method that cannot distinguish the intact conjugate from cleaved fragments reports the half-life of whatever the antibody or the mass filter happens to detect, which for an acylated or fused molecule may be the carrier rather than the payload. Before comparing two published half-lives, establish what each assay was counting. Most of the time that question is easier to answer than the half-life difference it explains.