Two-Stage Testing for Peptide Impurities

Evidence

Therapeutic peptides can acquire closely related variants during synthesis or degradation. Detecting those variants is an analytical problem; deciding whether one changes biological risk is a separate problem. Two independent studies address those stages from different directions. A fresh chromatography study tested how charge-based separation complements standard reversed-phase methods, while a full-text salmon-calcitonin study compared computational immune-risk predictions with laboratory assays.

The chromatography study, first published July 23, 2026, evaluated ion-exchange chromatography, or IEX, for therapeutic-peptide impurity profiling. The researchers tested four short columns: two strong cation exchangers and two strong anion exchangers, each in a 2.1-by-50-millimeter format. Their abstract says standard protein-oriented IEX methods did not work adequately for the target peptides. They then compared salt gradients at pH 2 for cation exchange and pH 10 for anion exchange with phosphate-citrate and polyamine-based pH gradients spanning pH 2.5 to 9. Salt-gradient elution performed better across the evaluated approaches.

Charge was not the only interaction governing the separation. The researchers reported substantial secondary hydrophobic interactions, which broadened peaks and reduced peptide recovery. Adding at least 30% acetonitrile mitigated those effects. They also used forced-degradation samples to benchmark the developed IEX methods against an in-house reversed-phase liquid chromatography purity method. Finally, they explored whether IEX retention could estimate peptide isoelectric points, reporting both promise and current limitations. Because only the abstract was ingested, it supports these method directions but not peptide-by-peptide resolution values, sample numbers, or complete forced-degradation results.

The independent 2024 study asked what can happen after a peptide impurity is identified. It examined 20 sequence variants associated with synthetic or degraded salmon calcitonin, including insertions, deletions, oxidation, acetylation, deamidation, and substitution. The team first used EpiMatrix to predict binding frames across nine HLA-DR supertypes and JanusMatrix to assess similarity to human proteins. The 20 impurity sequences did not differ significantly from the active pharmaceutical ingredient in overall EpiMatrix scores, although some modifications created predicted new epitopes. That result argues against treating every detected variant as equivalent evidence of added immune risk.

The researchers then tested selected variants with orthogonal experiments. Three impurities were compared with relevant salmon-calcitonin regions in competition assays spanning eight HLA-DRB1 alleles. Some predictions aligned with measured binding changes, while others did not. Four other impurities were evaluated in naïve T-cell assays. At equal peptide concentrations in cells from 16 donors, the active ingredient produced a positive response in 7 of 16 donors, or 44%. The four impurities produced responses in 9, 9, 10, and 11 donors, respectively. However, differences in response intensity versus the active ingredient were not statistically significant in the reported paired and unpaired nonparametric tests.

A second T-cell experiment used cells from 20 donors and added each impurity to the reference drug product at its observed abundance. Three of four impurities increased the incidence of donor responses relative to the reference product alone. This was an in vitro risk signal, not evidence of immune reactions in patients. The study's central contribution was methodological: sequence prediction, HLA binding, and donor-cell responses supplied different evidence, and direct comparison with the unmodified peptide helped keep each result relative rather than absolute.

Analysis — Detection and Consequence Need Different Tests

The cross-study pattern is a two-stage quality problem. The fresh IEX study shows that closely related therapeutic peptides may require a separation mode based on charge as a complement to reversed-phase chromatography, while also revealing that hydrophobic behavior can interfere even inside a nominally charge-based method. The salmon-calcitonin study shows that finding a sequence variant does not by itself establish biological consequence: overall computational scores, individual HLA-binding changes, and donor T-cell responses did not always move together. This is analysis across distinct peptide systems, not proof that a particular chromatographic peak is immunogenic.

Together, the studies support an emerging, unproven workflow in which orthogonal separation first expands which variants are visible, then sequence-resolved immune assays prioritize which variants deserve deeper investigation. The key connection is not that IEX predicts immunity; it does not. Rather, better physical profiling defines the candidates that biological assays can test. A rigorous bridge between the stages would require collecting individual IEX-resolved peaks, identifying each structure by mass spectrometry, measuring its abundance, and testing the same isolated variant against the parent peptide in matched immune assays.

Limitations

The two studies examined different peptides and did not share samples, so the proposed two-stage workflow is a synthesis rather than a directly validated pipeline. The 2026 IEX paper was available only as an abstract. It does not expose the complete peptide panel, chromatograms, impurity identities, numerical resolution, recovery values, replicate structure, or full forced-degradation comparisons. It therefore cannot establish that IEX is superior to reversed-phase chromatography generally or that its retention estimates are accurate enough to replace direct isoelectric-point measurements.

The salmon-calcitonin study supplied commercially reusable full text, but its immune experiments were in vitro and used selected impurities. The equal-concentration experiment tested impurities at far higher proportions than they would ordinarily contribute to a drug product, while the abundance-matched experiment still cannot reproduce dosing, antigen processing, or immune regulation in people. Donor cohorts were small, the active peptide itself was immunogenic in the assay, and several comparisons lacked statistical separation. Prediction disagreements also show that computational screening cannot substitute for experimental validation. Neither study establishes clinical safety, manufacturing specifications, or a universal impurity threshold. Stronger evidence would test structurally identified variants from the same peptide batch across complementary chromatographic, mass-spectrometric, cellular, and ultimately clinical measurements.