A conjugate is three parts — carrier, linker, payload — and the linker carries most of the design burden and nearly all of the analytical burden. The carrier decides where the construct distributes. The payload decides what chemistry happens once it is free. The linker decides whether the payload is still attached on arrival, and whether it ever comes off. A great deal of what can go wrong with a conjugate is a linker property.

One scoping note before any numbers. Nearly every quantitative result here was measured on antibody-drug conjugates. Peptide carriers differ in ways that matter: far fewer attachment sites, much faster clearance, and no comparable residence time in the endolysosomal pathway. The chemistry transfers more readily than the kinetics. Where a figure below comes from an antibody conjugate, this guide names that scope, and the figure should be read as evidence about that system rather than as a specification for a peptide one.

Cleavable and non-cleavable name a fate, not a bond strength

The distinction is not weak bond versus strong bond. Both classes are meant to survive circulation intact. What separates them is what has to happen before the payload becomes free.

A cleavable linker contains a designed trigger — a bond that some specific condition breaks much faster than the surrounding chemistry: a protease, a low pH, a reducing environment, a glycosidase. A non-cleavable linker contains no such trigger. The payload is liberated only when the carrier itself is degraded, and what is liberated is not the payload but the payload still bearing the linker and the carrier residue it was attached to.

That catabolite is a different molecule from the free payload, and its properties decide whether the design works at all. For maytansinoid antibody conjugates built on a non-cleavable SMCC linker, the catabolite is lysine-linked and charged. A triglycyl peptide linker (CX) was designed so that a significant fraction of its carboxylic acid catabolite could be uncharged in the acidic lysosome and so diffuse out into the cytosol. Across several cancer cell lines the two conjugate types had similar cytotoxic activity in vitro; in other lines, including a multidrug-resistant one, the CX conjugate was more active, with IC50 values 5- to 100-fold lower [1]. Keep the scope attached to that range: it was reported across cell lines, not as a fixed property of the linker.

Protease-sensitive designs, and what actually cleaves them

The archetype places a dipeptide — valine-citrulline is the canonical choice — between the carrier and a self-immolative spacer, usually a para-aminobenzyl carbamate. A protease hydrolyzes the amide on the far side of the dipeptide, and the exposed spacer then fragments on its own, releasing the payload unmodified. That spontaneous fragmentation step is textbook mechanism rather than a reported finding and is stated here without citation.

The early literature on these linkers is also a lesson in what a linker does to the rest of the molecule. In an early study of dipeptide-linked minor-groove-binder antibody conjugates, the initial adducts in the amino-CBI series, built with valine-citrulline attached to a para-aminobenzyl carbamate spacer, were completely aggregated. Removing the self-immolative spacer, switching to the more hydrophilic valine-lysine sequence, and inserting a tetraethylene glycol unit between the antibody and the peptide gave conjugates that were non-aggregated even with as many as eight drugs per antibody; a related construct released its payload on treatment with human cathepsin B [2]. Linker hydrophobicity is not a side effect. For a hydrophobic payload it is often the dominant solubility problem.

The trigger is also less specific than its name. The valine-citrulline linker was designed for cathepsin B, but suppressing cathepsin B expression by CRISPR-Cas9 deletion or shRNA knockdown had no effect on the efficacy of conjugates carrying a VC(S) linker and a monomethyl auristatin E payload, and mass spectrometry of payload release suggested other cysteine cathepsins can cleave that linker. A non-protease-cleavable enantiomer, VC(R), still killed cells effectively through a cysteine-VC(R)-monomethyl auristatin E catabolite generated by lysosomal catabolism. Only when the payload was switched to one that must be released to function — a pyrrolobenzodiazepine dimer that has to bind DNA — did the stereochemistry matter, and there the VC(S) conjugate was at least 20-fold more cytotoxic than the VC(R) [3]. Those authors drew a conclusion about the therapeutic index of protease-cleavable linkers; the narrower and more portable point is that a linker can have several routes to an active catabolite, so an activity readout is weak evidence that your trigger fired.

Cleavage you did not design

A cleavable bond is cleavable wherever the enzyme is. Working with an auristatin payload site-specifically attached through a valine-citrulline-para-aminobenzylcarbamate linker, one group reported a range of stabilities across various attachment sites on the same antibody, found that the conjugation site plays an important role in determining linker stability in mouse plasma, and showed that the cleavage seen in mouse plasma is not mediated by cathepsin B. The same work reported that this cleavage was not detected in primate plasma in vitro [4]. The enzyme behind it was later identified as carboxylesterase 1C, and small chemical modifications to the linker were shown to modulate that esterase's activity toward it without significantly altering intracellular processing by cathepsin B [5].

The bench consequence, stated as reasoning rather than as a finding those authors reported: a linker stability result is a statement about a matrix and a species, and a construct stable in one plasma can be hydrolyzed in another by an enzyme the design never considered.

pH-sensitive designs and what they trade

Acid-labile chemistry — the hydrazone is the standard example — exploits the pH difference between circulation and the endolysosomal compartment. It is a rate ratio, not a switch. The same bond that hydrolyzes quickly at pH 5 hydrolyzes slowly at pH 7.4.

A mechanistic modeling study of doxorubicin attached to block copolymer micelles through acid-labile hydrazone bonds, comparing three drug-binding linkers (hydrazide, aminobenzoate-hydrazide and glycine-hydrazide), reported biphasic release kinetics for all three formulations and found the extent of drug release at specific time intervals greater at pH 5.0 than at pH 7.4, in the order hydrazide, then aminobenzoate-hydrazide, then glycine-hydrazide. The model only quantified the observed kinetics once it included terms for unconjugated drug and for two distinct populations of conjugated drug, and it correlated changes in release kinetics with instability of the hydrazone conjugation during prolonged storage [6]. That is a polymer micelle system with one payload and three linkers, not a general result about hydrazones; but two features generalize as warnings. A release curve that needs more than one conjugated population to fit is telling you the material is heterogeneous, and a linker that leaks slowly on the shelf will report a different release profile after storage than it did fresh.

An enzyme-dependent trigger identifies a compartment by a different property than a pH-dependent one does, and whether it is the more selective handle in a given system is something to measure rather than assume. In a pyrrolobenzodiazepine dimer series compared against the corresponding free-imine conjugates, a beta-glucuronidase-cleavable cap at the N10 position gave conjugates whose activity depended on beta-glucuronidase, demonstrated through CRISPR-Cas9 knockdown and through addition of the exogenous enzyme. That glucuronide series was also reported to show better serum stability and improved conjugation efficiency, and to reach a high drug-to-antibody ratio without aggregation [7].

The attachment bond is part of the linker

Whatever the trigger, the bond that joins the linker to the carrier is itself a stability liability, and for thiol conjugation it is the usual one. Maleimide-thiol adducts undergo two competing reactions in plasma: elimination of the maleimide by a retro-Michael reaction, which loses the drug-linker from the conjugate entirely, and hydrolysis of the thiosuccinimide ring, which yields a derivative resistant to that elimination. A drug-linker incorporating diaminopropionic acid places a basic amino group adjacent to the maleimide, positioned to catalyze ring hydrolysis intramolecularly; that group induces rapid hydrolysis at neutral pH and room temperature, and once hydrolyzed the drug-linker is no longer subject to maleimide elimination [8].

Note the direction: you want hydrolysis to win that race. And retro-Michael loss is not payload release. The payload departs with its linker attached, so an assay for free payload will not see it, while an assay for conjugated payload will.

Site of attachment is a variable in its own right, not a detail. Beyond the mouse-plasma stability range across sites already described [4], a study developing pharmacokinetic assays for disulfide-linked, self-immolating conjugates found that the conjugation site of the linker-drug on the antibody affected the kinetics of drug release, and listed tertiary structure, fractional solvent accessibility, the pKa of the conjugation site, the surrounding residues' isoelectric point and electrostatic charge as characteristics that may directly impact release kinetics [9]. That list is the authors' set of candidate contributors, offered with a modal; it is not a ranked or validated model.

Drug-to-peptide ratio is a mean over a distribution

The average ratio is total moles of payload divided by total moles of carrier. The denominator is all carrier, including carrier molecules that took up nothing. Two preparations with the same average can be entirely different materials.

Take three hypothetical mixtures of a carrier bearing 0, 2, 4, 6 or 8 payloads. Mixture A is half at 2 and half at 6: 0.5 x 2 + 0.5 x 6 = 4. Mixture B is entirely at 4. Mixture C is half at 0 and half at 8: 0.5 x 0 + 0.5 x 8 = 4. All three report an average ratio of 4; in C, half the carrier carries nothing and the rest carries the maximum. The number is the same and the material is not.

Loading is not a free parameter either. Coupling monomethyl auristatin E to the interchain-disulfide cysteines of an anti-CD30 antibody and purifying by hydrophobic interaction chromatography gave populations with two, four and eight drugs per antibody. In vitro potency tracked loading directly, with IC50 values ordered E8 below E4 below E2, while plasma clearance in mice rose with loading: the eight-loaded species cleared 3-fold faster than the four-loaded and 5-fold faster than the two-loaded [10]. If both of those are ratios of the same clearance parameter, then the four-loaded cleared about 1.7-fold faster than the two-loaded — arithmetic on their two figures, not a value they report.

For peptide carriers the count is usually small, often one, so heterogeneity shows up less as a spread in count than as positional isomers: which of the few available nucleophiles reacted. A peptide with a free N-terminal amine and a single lysine side chain presents two competing amines of different pKa, and an amine-reactive reagent will give both mono-substituted products plus the doubly substituted one. Nothing in the average ratio distinguishes them. Positional isomerism of this kind has been investigated by molecular modeling for the lysine-linked conjugate trastuzumab emtansine [11].

Measuring the ratio and the distribution

For cysteine conjugates, hydrophobic interaction chromatography is the standard technique for drug distribution, naked antibody content and average ratio, because each added payload contributes hydrophobic surface. Lysine conjugates are not amenable to it, because of their higher heterogeneity; for those, denaturing mass spectrometry and UV-visible spectroscopy have been the most powerful approaches. Native mass spectrometry, combined with high-resolution measurement or charge reduction, was reported to give accurate average ratios and drug load profiles for trastuzumab emtansine, and native ion mobility mass spectrometry was used to determine distribution profiles and average ratios directly [11]. A broader review of liquid chromatography for these conjugates covers hydrophobic interaction, reversed-phase, size-exclusion and multidimensional separations for evaluating drug load distribution, determining the average ratio, and analyzing process- and storage-related impurities [12].

Two habits are worth carrying over to peptide conjugates, stated as practice rather than findings: report the distribution alongside the mean, and report which method produced it, because a chromatographic average and a mass-spectrometric average can disagree.

The distribution can also be narrowed at the source rather than by purification. In one site-specific approach, the native glycans at Asn-297 were remodeled enzymatically with galactosyl- and sialyltransferases to introduce terminal sialic acids, those were oxidized with periodate to aldehydes, and aminooxy-functionalized payloads were attached by oxime ligation; hydrophobic interaction chromatography and LC-MS showed incorporation of about 1.6 payloads per antibody, approximating the number of sialic acid residues [13]. The strategy carries — fix the number of reactive handles and the distribution follows — but the figure does not, being specific to that glycan chemistry.

Stability testing of the linkage

A stability study has to specify, at minimum, the matrix, the species it came from, temperature, duration, and which of three quantities is tracked: total carrier, conjugated payload, and free payload. Those three trend differently, and a design can fail on one while looking sound on another. This is convention rather than a published standard, so it is asserted without citation.

The assay itself often contains release chemistry, which is easy to forget when reading a number out of it. For conjugates built on enzymatically cleavable peptide linkers such as maleimidocaproyl-valine-citrulline-para-aminobenzyl, conjugated-drug pharmacokinetic assay strategies involve cleavage with cathepsin B or papain to liberate and measure the antibody-conjugated payload; for disulfide-linked, self-immolating conjugates one group instead used immunoaffinity capture followed by chemical reduction at basic pH, self-immolation to liberate the active drug, and quantitation by multiple reaction monitoring LC-MS/MS, reporting accuracy and precision within 20 percent [9]. Whatever the format, the reported quantity is defined by the release step, so two laboratories using different release chemistries are not reporting the same number.

What is convention and what is not known

  • There is no accepted numerical threshold for how much in-matrix cleavage makes a linker unacceptable. Comparisons in the literature are almost always against a comparator construct in the same experiment, not against an absolute criterion.
  • Whether stability in one matrix predicts stability in another is not established in general. The mouse carboxylesterase result is one published counter-example, and it was specific to valine-citrulline-para-aminobenzylcarbamate linkers in mouse plasma [4, 5].
  • Demonstrating that a linker is cleavable is not evidence that cleavage is what released the payload in a given experiment. Multiple catabolic routes to an active species have been shown for a valine-citrulline auristatin conjugate [3].
  • Acid-lability is a rate ratio. A design that releases faster at low pH still releases at neutral pH, and how much slower is a measured quantity for a specific construct, not a class property.
  • The quantitative record for peptide carriers specifically — as opposed to antibody carriers — is thin. Treat linker chemistry as transferable and linker kinetics as something to re-measure.