Racemization and Epimerization During Peptide Synthesis

A synthesized peptide can carry the correct sequence, the correct intact mass, and a single sharp peak on a reversed-phase chromatogram, and still contain a population of chains in which one residue has the wrong handedness. Those chains are diastereomers of the target. They form during the synthesis itself, mostly at the moment a carboxyl group is activated, and they are the impurity class the two most common release assays — intact mass and area-percent purity — are least equipped to find. This guide covers the chemistry that generates them, what makes it worse, the per-chain arithmetic, and the methods that resolve them.

One note on words. Racemization strictly means drift toward a 50:50 mixture of enantiomers; epimerization means inversion at one stereocenter among several, which is what actually happens in a growing chain. The literature uses both loosely and often interchangeably, and so does this guide.

Mechanism one: the 5(4H)-oxazolone

When the carboxyl group of an N-acylated amino acid is activated — as an O-acylisourea from a carbodiimide, as an active ester from an onium salt, as a mixed anhydride — the carbonyl oxygen of the preceding amide bond can attack that activated carbon intramolecularly and displace the leaving group. The product is a five-membered 5(4H)-oxazolone, or azlactone. Its α-carbon now sits between a carbonyl and a C=N, and the proton on it is acidic enough for a mild tertiary amine to remove. The resulting anion is planar and achiral, and reprotonation occurs on either face. Because the oxazolone is itself still activated, it goes on to react with the incoming amine and delivers a mixture of configurations at that center.

Two features of this route drive most of the practical advice around it. First, the oxazolone forms from the activated carboxyl component, so anything that shortens that species' lifetime before aminolysis limits the damage. Second, additives can intervene at two separate points. Miyazawa and colleagues, coupling Z-Gly-L-Val-OH with H-L-Val-OMe in DMF with a carbodiimide, attributed the elimination of racemization they saw when 1-hydroxybenzotriazole and copper(II) chloride were used together to two distinct effects: HOBt lowers the tendency of the activated forms to close to the 5(4H)-oxazolone at all, and copper(II) chloride suppresses racemization of the small amount that forms anyway ^[6]. That is the authors' reading of their own model couplings rather than a general law, but it explains why HOBt-class additives became standard.

The oxazolone route also has a consequence for measurement. Benoiton and colleagues examined aminolysis of preformed 5(4H)-oxazolones by amino acid benzyl esters and found substantial asymmetric induction: the incoming nucleophile does not attack the two faces of the planar intermediate equally, and diastereomeric excesses reached −32 and +53 depending on the N-acyl group, the side chain of the activated residue, the aminolyzing nucleophile, the solvent, and the temperature ^[5]. For their N-acyl series they reported that the epimeric content of the products did not always correctly reflect the relative tendencies of the derivatives to racemize ^[5]. A measured D-content is a property of the whole coupling system, not a clean readout of one residue's tendency to lose its proton.

Mechanism two: direct enolization

The second route needs no oxazolone. A base abstracts the α-proton directly from the activated derivative, giving an enol or enolate that is planar at the α-carbon; reprotonation is again non-selective. It is favored when the side chain stabilizes the resulting carbanion, which is why residues bearing an anion-stabilizing group close to the α-carbon dominate the list of offenders. Both routes are base-driven and both act on the activated species, so they respond to many of the same knobs, and a single measured D-isomer percentage rarely tells you which one was operating.

Which residues are worst

Cysteine. The most quantitatively documented case. Han, Albericio and Barany used a model system that resolves H-Gly-L-Cys-Phe-NH₂ from H-Gly-D-Cys-Phe-NH₂ by HPLC and found that standard protocols — phosphonium and aminium salts such as BOP, HBTU, HATU and PyAOP, with HOBt or HOAt and a tertiary amine base such as DIEA or NMM, using a 5-minute preactivation — gave D:L ratios in their model peptide across the range 5–33% ^[1]. They described this as contrary to the conventional wisdom of the field at the time.

Histidine. The imidazole side chain makes histidine a recurring problem. In a model 20-mer containing all 20 natural amino acids assembled under microwave heating, Palasek and colleagues found cysteine, histidine and aspartic acid to be the residues susceptible to racemization ^[3]. Hojo and colleagues, developing an aqueous microwave-assisted protocol with a triazine-based coupling reagent, reported that coupling could be performed with low levels of histidine racemization in their water-based system, and applied it to a histidine-containing hexapeptide ^[4]. That is a claim about their system, not a general statement that aqueous conditions are safer.

Aspartic acid. Aspartate has a second, indirect route. Base can close the side-chain carboxyl onto the backbone to give an aspartimide, and ring opening of that succinimide regenerates a mixture containing both the inverted residue and the β-linked isoaspartyl isomer. Palasek and colleagues reduced aspartimide formation, and the aspartic acid racemization that follows it, by adding HOBt to the deprotection solution and/or using piperazine in place of piperidine ^[3]. Separately, Zhang and colleagues state that aspartic acid epimerization is faster than that of other amino acids and can be accelerated by free radicals and peroxides; in their model peptide it was rapid when ascorbic acid and Cu²⁺ coexisted, and its extent correlated positively with hydrogen peroxide generation ^[9].

Serine. Serine shows up in real impurity profiles. Characterizing low-level isomeric impurities in a commercial synthetic peptide, Zhang and colleagues identified D-Ser8, D-His1 and D-Asp9 species ^[10]. Two of those three are the residues the synthesis literature already flags; serine is the reminder that the list is not closed.

Glycine is the trivial exception: no α-stereocenter, so it cannot epimerize. Beyond these cases the ranking of the remaining residues is far less firmly established, and much of what circulates as a hierarchy is inherited convention rather than a measured series.

Which conditions are worst

Preactivation.

Preactivation. The largest quantified effect in the cysteine study was the preactivation step: Han and colleagues reported racemization levels reduced in general by a factor of 6- or 7-fold simply by omitting it, so the amine is present when the activated species forms ^[1]. Every minute an activated ester sits in base without a nucleophile is time available for oxazolone closure.

Base identity and quantity. The same study found 2,4,6-trimethylpyridine (collidine) substantially better than DIEA or NMM, that halving the amount of base helped, and that switching from neat DMF to CH₂Cl₂–DMF (1:1) helped ^[1]. Their combined recommendations reached under 1% racemization per step for cysteine in Fmoc synthesis ^[1]. Palasek and colleagues likewise found that using collidine in the coupling minimized formation of D-cysteine ^[3].

Protecting group. The thiol protecting group is not neutral, and the weak-base workaround above costs something. Hibino and colleagues reported that the weak-base and carbodiimide methods are appreciably inferior in coupling efficiency to the standard phosphonium/uronium plus DIEA protocol, and that for S-trityl cysteine the weak-base route could not reach an acceptable level of under 1.0% even with preactivation omitted; the 4,4'-dimethoxydiphenylmethyl and 4-methoxybenzyloxymethyl groups did reach that level under the standard protocol, and reduced the rate of racemization relative to trityl even under high-temperature microwave-assisted synthesis ^[2].

Temperature. Heating accelerates the same base-mediated chemistry. Palasek and colleagues found that lowering the microwave coupling temperature from 80 °C to 50 °C limited racemization of histidine and cysteine, and that in their hands racemization during coupling was limited to the activated ester state up to 80 °C — letting those two residues be coupled conventionally while the rest of the chain was assembled under microwave heating ^[3].

Segment condensation. Coupling two fragments activates the C-terminal residue of a fragment that already carries an N-acyl group — exactly the substrate the oxazolone route wants — and the numbers reflect it. In a convergent 5+4 synthesis of a nonapeptide, Xaus and colleagues monitored racemization of the acyl-activated amino acid by HPLC and reported that DIC/HOBt and BOP/HOBt/NMM always gave highly optically impure products containing 10–20% of the D-epimer ^[7]. N-methylated residues are a related liability: N-methylated residues are a related liability: Nishiyama and colleagues reported that using N-hydroxysuccinimide and CuCl₂ together with a succinimidyl uronium coupling reagent eliminated racemization of the carboxy-terminal N-methylamino acid residue during segment condensation ^[8]. The mechanism for N-alkyl residues is less settled than the oxazolone route, but the liability is documented.

Per-step numbers do not stay small

Racemization is reported per coupling, and a per-coupling number understates what reaches the vial. Treat the steps as independent — an approximation, not a measured fact for any particular sequence — and the arithmetic is a simple binomial.

Take a 30-residue peptide with 0.5% inversion at each of 30 couplings. The fraction of chains surviving with every center intact is 0.995³⁰ = 0.8604, so 86.0% of chains are all-L and 14.0% carry at least one inverted residue. Chains with exactly one inverted residue account for 30 × 0.005 × 0.995²⁹ = 0.1297, or 13.0%; the remaining roughly 1.0% carry two or more. That 13.0% is not one impurity. It is spread across 30 distinct single-epimer species, averaging 0.43% each — small enough that most fall below a 0.5% integration threshold individually while summing to an eighth of the material.

Now double the per-step rate to 1.0%. The all-L fraction becomes 0.99³⁰ = 0.7397, so 74.0% of chains are clean and 26.0% carry at least one inverted center. Note the denominator: doubling the per-step rate did not double the affected fraction. It went from 14.0% to 26.0%, a factor of about 1.86, because the two quantities are related exponentially rather than linearly. The corollary runs the other way too — the difference between 0.5% and 0.2% per step is not cosmetic across a long chain.

Why intact mass cannot see any of this

An inverted residue changes no atom count and no bond count. The epimer has the identical molecular formula and therefore the identical monoisotopic and average mass as the target. Zhang and colleagues put it plainly when introducing a chiral-phase method: the modification causes no change in protein m/z and may be overlooked during protein analysis ^[9]. Berthias and colleagues extend the same logic to sequence-level workflows, noting that unguided proteomic algorithms based on peptide masses are oblivious to D-amino-acid-containing peptides ^[12].

Tandem MS does not fix this on its own: backbone fragmentation of an epimer yields fragments themselves isobaric with the target's. Fragment intensities can differ, but intensity differences are instrument- and condition-dependent, and a mass spectrum alone cannot assign configuration. Something orthogonal to mass has to do the separating.

How D-isomers are actually detected

Four approaches appear in the literature; they answer different questions.

• Reversed-phase separation against authentic diastereomer standards. Because an epimer is a diastereomer, it is in principle separable on an achiral column. This is how the cysteine work was quantified: HPLC resolution of H-Gly-L-Cys-Phe-NH₂ from H-Gly-D-Cys-Phe-NH₂ in a deliberately simple model system ^[1]. Quantitative and cheap, but it requires the epimer in hand and a pair that resolves.
• Chiral stationary phase on the intact peptide. Zhang and colleagues validated a site-specific chiral-stationary-phase HPLC method for aspartic acid epimerization in a model active-site peptide, with linearity from 1 to 200 µg/mL and recoveries between 85% and 115% at the limit of quantification and at low, medium and high concentrations ^[9]. This gives configuration directly, without hydrolysis.
• Hydrolysis followed by chiral amino acid analysis. The workhorse for identifying which residue inverted. In the commercial-peptide impurity study, targeted isomeric species were collected offline from a reversed-phase separation, lyophilized, hydrolyzed in deuterated hydrochloric acid with substrates to suppress D/L shifting, chirally derivatized, and analyzed by RP-UPLC tandem MS against standards ^[10]. Deuterated acid makes that step interpretable, because hydrolysis itself racemizes: Péter and colleagues, evaluating microwave-driven hydrolysis of peptide samples in 6 N deuterium chloride, determined the racemization introduced there from the amount of deuterium incorporated, and reported that high recovery of sensitive amino acids with minimal racemization was obtained using an intermediate irradiation power and a 30-minute reaction time ^[13]. A chiral amino acid result that does not account for hydrolysis-induced racemization reports a number that includes the sample preparation.
• Ion mobility. A gas-phase option needing no chiral column. Berthias and colleagues applied high-resolution differential ion mobility (FAIMS) to d/l peptide epimers and fully separated all 18 pairs of representative epimer species examined, spanning masses of roughly 400–5,000 Da and charge states of 1–6; the orthogonality of MS to high-resolution FAIMS exceeded that to linear IMS by 6×, and on the basis of constant isomer resolution across those ranges they projected that the approach should extend to larger species ^[12].

The scale these methods have to cope with is worth stating. Du and colleagues developed a high-throughput HPLC-MS/MS strategy profiling all 20 isobaric epimers of a β-amyloid peptide arising from Asp, isoAsp and Ser isomerization, allowing direct detection and identification of all possible stereoisomers of those residues ^[11]. One peptide, twenty isobaric species that an intact-mass measurement reports as one compound.

What is measured and what is convention

Everything above with a number attached is a measurement made in a named model system. What follows is not. There is no general, transferable table of racemization propensity per residue holding across coupling reagents, bases, solvents, temperatures and neighboring sequence — published numbers are tied to the model systems that produced them, and Benoiton's finding that product epimeric content did not always track racemization tendency warns directly against reading one as the other ^[5]. Whether a long sequence racemizes independently at each step, as the arithmetic above assumes, is not established for any particular chain. And the standard advice to install sensitive residues at reduced temperature extrapolates from the temperature dependence that was measured; it is not a validated rule for arbitrary sequences.

What this means for reading a document

A certificate of analysis showing a reversed-phase area percent and an intact mass contains no stereochemical information. That is a statement about what those two measurements are, not an accusation about any material. If stereochemical purity matters for an experiment, the question is which orthogonal method was run: a chiral stationary phase on the intact peptide, a hydrolysis-plus-chiral-amino-acid workflow with a stated control for hydrolysis-induced racemization, or a separation against authentic diastereomer standards. If none was run, the honest reading is that D-isomer content is unmeasured — not that it is zero.