A peptide sitting in a vial is not doing nothing. Three distinct chemistries run on it at all times, at rates set by its sequence, solvent, pH, temperature, and water content. Hydrolysis cuts the backbone. Deamidation converts side-chain amides to acids and, on the way, rearranges the backbone itself. Oxidation adds oxygen to sulfur and to aromatic rings. Each pathway has a different residue preference, a different pH dependence, and a different analytical signature — which is the practical point, because the signature is how you find out whether the material on your bench is what the label says.

What follows covers the chemistry of each route, what pH and temperature do to it, and how each shows up analytically — separating what has been measured from what is convention.

Hydrolysis: cutting the backbone

The amide bond that links residues is thermodynamically unstable to hydrolysis but kinetically slow — which is why proteins exist at all. In practice, backbone cleavage in a stored peptide is rarely uniform along the chain. It concentrates at a small number of positions, and the best-characterized of these is aspartate.

Oliyai and Borchardt used the model hexapeptide Val-Tyr-Pro-Asp-Gly-Ala to map this in detail. At the very acidic pH values they sampled — 0.3, 1.1, 1.5, 2.0 and 3.0 — degradation followed pseudo-first-order kinetics, and in that region the peptide predominantly underwent specific acid-catalyzed hydrolysis of the Asp-Gly amide bond, producing the tetrapeptide Val-Tyr-Pro-Asp and the dipeptide Gly-Ala. Their pH-rate profiles, which spanned pH 0.3 to 5.0, showed that the ionization state of the aspartate side-chain carboxyl significantly altered the rate for both this hydrolysis and the competing cyclic-imide route, with the ionized form being the more reactive. They observed little or no buffer catalysis for either pathway [1].

Aspartate is not the only site. Tyler-Cross and Schirch found that every pentapeptide in their series with a serine following an asparagine underwent cleavage at neutral and alkaline pH, yielding a tripeptide and a dipeptide, at about ten percent of the deamidation rate in the same peptides [2]. Goolcharran and colleagues, working on Gly-Gln-Asn-X-His pentapeptides, reported that one member of that series, Gly-Gln-Asn-His-His, also underwent cleavage at the Asn-His bond itself, and attributed the enhanced rate of that cleavage to general base behavior of the histidine side chain increasing the nucleophilicity of the asparagine side-chain amide [3].

The generalization worth carrying is narrow but useful: backbone cleavage tends to be neighbor-assisted, with a side chain adjacent to the scissile bond supplying the nucleophile or the catalysis. A peptide with no Asp, no Asn-Ser and no Asn-His is not thereby immune — only slower.

Deamidation: the succinimide route

Deamidation is the pathway most likely to change your peptide without changing anything you would notice by eye. Asparagine's side-chain amide is attacked by the backbone nitrogen of the following residue, forming a five-membered cyclic imide (a succinimide) and releasing ammonia. That ring then opens by hydrolysis at either of two carbonyls, giving normal aspartate or isoaspartate — the latter with an extra methylene in the backbone and the original α-carboxyl now hanging off as a side chain. Three features of this mechanism drive everything downstream.

The following residue dominates. Stephenson and Clarke measured succinimide formation at pH 7.4 in Val-Tyr-Pro-X-Y-Ala peptides. Asparaginyl peptides formed succinimide 13.1 to 35.6 times faster than the corresponding aspartyl peptides; glycine at the following position gave rates 6.5 to 17.6 times faster than alanine, and serine 1.6 to 4.5 times faster than alanine. The overall range across their set was 232-fold, which they took to suggest that sequence can be an important determinant of stability in flexible peptides. They were careful to add that in proteins the range may be much larger, because specific conformations may greatly enhance or inhibit the reaction [4]. Tyler-Cross and Schirch reached the same conclusion from the other direction: in their pentapeptides, in neutral and alkaline solution, the residue on the amino side of the asparagine had little or no effect regardless of charge or size, while increasing size and branching on the carboxyl side slowed deamidation by as much as 70-fold relative to the Asn-Gly sequence, which was fastest [2].

The products are mixed, and the mixture depends on pH. Tyler-Cross and Schirch found isoaspartate and aspartate products in roughly a 3:1 ratio at neutral and alkaline pH. In acidic solution, the aspartate peptide was the only deamidation product they detected, and the rate became insensitive to sequence [2]. Oliyai and Borchardt saw the complementary picture from an aspartyl starting material: at pH 4.0 and 5.0 the hexapeptide both isomerized via the cyclic imide and underwent Asp-X hydrolysis, while above pH 6.0 the isoaspartyl peptide was the only observed product [1].

Histidine is a special case. Goolcharran and colleagues showed that a histidine immediately after the asparagine does two opposing things. Below pH 6.5 it accelerated deamidation, which they proposed as general acid catalysis stabilizing the oxyanionic transition state of imide formation. Above pH 6.5 the bulk of the histidine side chain apparently restricted flexibility around the reaction site and slowed the reaction relative to glycine — while still acting as a catalyst relative to valine. A histidine two residues away had little or no effect [3].

Glutamine deamidates too

Glutamine is usually treated as the slow one, on the reasoning that its analogous cyclic intermediate is one ring atom larger than asparagine's succinimide and therefore harder to form. That reasoning is structural rather than measured, and it does not hold everywhere. Joshi and Kirsch degraded glucagon fragment 22-29 (FVQWLMNT) at 60 °C in dilute hydrochloric acid or phosphate buffer over pH 1 to 3 and identified nine major degradation products, including both deamidated forms and a cyclic imide intermediate. Under those acidic conditions they observed more facile deamidation at the glutamine than at the asparagine. They proposed that the asparagine's lower reactivity may be due to decreased electrophilicity of its side-chain carbonyl imparted by a parallel cleavage pathway at that residue [5]. That is a proposal about one peptide under acid, not a general reversal — but it is a useful warning that the Asn-fast/Gln-slow heuristic is pH-conditional.

Oxidation: sulfur first

Methionine is the most readily oxidized common residue, converting to methionine sulfoxide with a single oxygen addition of 15.9949 Da. Cysteine thiols oxidize and also scramble into disulfides; tryptophan, tyrosine and histidine oxidize under photochemical and radical conditions.

What governs which methionine goes first is largely geometric. Duenas and colleagues compared light-induced and chemical oxidation of recombinant human VEGF, which carries six methionines, analyzing tryptic digests by reversed-phase HPLC. Two weeks of fluorescent light exposure at 2 × 10⁴ lux gave 8 to 40 percent oxidation across all six residues. Hydrogen peroxide, the smaller of the two chemical oxidants, drove Met3, Met18 and Met55 to complete oxidation, while the bulkier tert-butyl hydroperoxide reached only 40 to 60 percent at those sites. Their conclusion was that methionines with exposed surface areas greater than 65 Ų and sulfur surface areas greater than 16 Ų were the most susceptible, and that the ability of the two chemical oxidants to react with methionine was governed by solvent accessibility of the methionine residues and steric limitations of the oxidizing agent — while photooxidation was instead dependent on the amount of energy imparted to the protein. Notably, far-UV and near-UV circular dichroism showed no secondary or tertiary structural change after chemical oxidation [6].

For a short unstructured peptide this accessibility argument mostly collapses — every methionine is exposed — so oxidation there should track oxidant exposure and headspace rather than sequence context.

Dry is not the same as stopped

Lyophilization slows these reactions; it does not switch them off, and the residual water content is the variable that matters. Strickley and Anderson studied human insulin lyophilized from pH 2 to 5 solutions and held at 35 °C across a range of relative humidities, measuring water content by Karl Fischer titration. Degradation rate increased with water content at hydration levels well below the suspected glass transition, and approached the solution rate at the 20 to 50 percent water content required to induce that transition. The solid-state apparent-pH-rate profiles ran parallel to the solution profile, and aniline trapping established that between pH 3 and 5 the rate-limiting step was intramolecular attack of the C-terminal AsnA21 carboxyl on its own side-chain amide carbonyl to form a cyclic anhydride [7].

Their follow-up makes the more interesting point: what the reactive intermediate becomes depends on mobility. In the glassy state at low water content, the intermediate reacted predominantly with water to give the deamidated product, with covalent dimer accounting for no more than about 15 percent of the total degradation. Raising the water content lowered the glass transition below 35 °C and progressively favored dimerization instead. Incorporating the material into a glassy trehalose matrix almost totally suppressed covalent dimerization, which they attributed to minimized molecular mobility and physical separation of the molecules [8].

The transferable lesson is narrower than the idea that dry means stopped. Strickley and Anderson describe the intramolecular cyclization at AsnA21 as needing only short-range conformational flexibility, and therefore as only modestly restricted even in the glassy state, while the competing bimolecular steps — the anhydride intermediate combining with either water or another insulin molecule — depend on the mobility of the reactants. Water is one of those two bimolecular partners, and in their glassy lyophiles it was the one that won: deamidation, not dimer, was the dominant fate of the intermediate [8]. Dryness throttles the routes that need whole molecules to move; on this evidence it does not put the chemistry beyond reach of the water that remains.

What temperature does, and what it does not tell you

Temperature accelerates all of these pathways, but the activation energies differ, so the product ratios at one temperature are not the ratios at another. This is the central weakness of accelerated stability testing on peptides: raising temperature can change which pathway dominates, not merely how fast the dominant one runs.

The common bench heuristic — that rates roughly double per 10 °C — is a rule of thumb, not a measurement, and it is worth seeing what it implies. Moving from 25 °C to −20 °C is a 45 °C drop, or 4.5 ten-degree steps, giving 2^4.5 ≈ 22.6-fold slowing. That is a real benefit and a modest one; it is not the thousand-fold that "frozen" suggests intuitively. Any given peptide's true factor requires its own measured Arrhenius data.

How each pathway shows up analytically

Reversed-phase HPLC is the first-line separation, and each pathway shifts retention characteristically — methionine sulfoxide is more polar than methionine and elutes earlier; deamidation adds a negative charge and shifts retention as well. But retention alone identifies nothing. The mass shifts do the work, and the arithmetic is worth having in front of you.

Deamidation is +0.9840 Da. Replacing the side-chain NH₂ with OH swaps 14.003074 + 2(1.007825) = 16.018724 for 15.994915 + 1.007825 = 17.002740, a difference of 0.984016 Da. This is a small number in the wrong place. The ¹³C–¹²C spacing is 1.00335 Da, so the deamidated monoisotopic peak sits only 1.00335 − 0.984016 = 0.019334 Da below the first isotope peak of the unmodified species. Resolving those two at m/z 1000 requires a resolving power of roughly 1000 / 0.019334 ≈ 51,700. On a 3,000 Da peptide, the deamidation shift itself is 0.984016 / 3000 = 328 ppm — comfortable for accurate-mass work, but only once you have separated it from the isotope envelope.

Methionine sulfoxide announces itself in CID. Guan and colleagues showed that low-energy collision-induced dissociation of a sulfoxide-containing peptide gives a characteristic neutral loss of methanesulfenic acid, CH₃SOH — 12.000 + 4(1.00783) + 15.99491 + 31.97207 = 63.998 Da. Because that side chain is labile under CID, they used electron capture dissociation to localize the modification, which produced extensive backbone fragmentation without detaching it. On atrial natriuretic peptide, a disulfide-containing peptide where CID was dominated by the 64 Da loss, ECD gave sequence-informative products including cleavages inside the disulfide-bonded ring, allowing direct localization without the usual reduction and alkylation steps [9].

Isoaspartate is the hard one. It is isobaric with aspartate — identical mass, no shift to look for — and it can co-elute. Two orthogonal handles exist. O'Connor and colleagues showed that electron transfer dissociation produces diagnostic c + 57 and z − 57 ions that define both the presence and the position of an isoaspartyl residue, and that these peaks are relatively abundant; they noted that the corresponding aspartate diagnostic ion is harder to detect on a lower-resolution ion trap because of interference from arginine side-chain fragments [10]. Separately, Ni and colleagues exploited the fact that the endoprotease Asp-N cleaves aspartyl but not isoaspartyl peptides, using it to differentiate and to enrich, then confirming site assignment by ETD. They detected as little as 0.5 percent isoaspartate in synthetic peptides assumed to be free of it [11].

The measurement can create the modification

This is the failure mode that most often produces a wrong answer, because it looks exactly like a real result. Zhang and Højrup reported a −17 Da modification corresponding to N-terminal cyclization in peptides carrying an N-terminal X-Asn-Gly motif, occurring spontaneously during sample preparation for bottom-up proteomics. Using synthetic peptides they confirmed that the ring forms between the N-terminus and its neighboring asparagine, and that the reaction shares the same succinimide intermediate as asparagine deamidation. Unlike the well-known N-terminal glutamine cyclization, this one was dependent on pH rather than ammonium concentration [12].

The practical implication is that a digest held at the wrong pH for the wrong length of time can generate succinimide-pathway chemistry in the tube and report it as a property of the sample. Blanks carried through identical handling are the practical control here: without one, a modification generated in the tube and a modification present in the sample look the same.

What is not known

A good deal of what circulates as peptide stability guidance is convention rather than measurement, and it is worth naming which parts.

  • Sequence-based deamidation prediction is calibrated on short flexible peptides. The rate tables come from pentapeptide and hexapeptide model systems, and Stephenson and Clarke explicitly flagged that conformation in a folded structure may enhance or inhibit the reaction well outside that range [4]. For a structured or cyclic peptide, the model-peptide half-life is a starting point, not a prediction.
  • Most catalog peptides have no published stability study at all. Rate data exist for a handful of well-studied model sequences. For the majority of research peptides, the pH of maximum stability, the dominant pathway, and the temperature dependence are simply not in the literature.
  • Storage recommendations are largely inherited, not derived. Standard practice — dry, cold, dark, desiccated, single-use aliquots, minimal freeze-thaw — is sound and consistent with the chemistry above, but the specific temperatures and shelf lives quoted for individual peptides are usually convention rather than the output of a measured stability program.
  • Oxidation depends on variables that are rarely recorded. Headspace oxygen, trace metals in the buffer, and light exposure history all matter, and almost none of it appears in a typical sample record.

The honest summary is that the mechanisms are well established and the rates usually are not. That asymmetry is why the analytical section above matters more than any storage table: for a given lot, measuring what happened beats predicting what should have.