CQA Assessment Using ChatGPT

Written By Simon Letarte

Recently I had the chance to play with ChatGPT to perform CQA risk assessment. It is something I attempted to do a few months ago when I was at a company. We could only use the most basic version of Copilot and the results were not entirely satisfactory. This time I used ChatGPT+ and the sequence for trastuzumab as an example. The results are scarily good. Notice it finds that W35 is not part of the CDR according to Kabat notation, but knows that it is adjacent to it and therefore oxidation of that residue could still disrupt binding.

Note that I did not ask to provide the literature references to this example, but one can easily add this to the prompt and ChatGPT will oblige. I tried several other modifications and was frustratingly surprised by the answers. Finally, I am not claiming that ChatGPT is better than Copilot for that type of work, just highlighting how fast the field has progressed in just a few months.

Prompt:

This monoclonal antibody (IgG1) is formed of 2 identical light chains and two identical heavy chains as follows:

>Anti-HER2 Light chain (1 and 2)
DIQMTQSPSSLSASVGDRVTITCRASQDVNTAVAWYQQKPGKAPKLLIYSASFLYSGVPS
RFSGSRSGTDFTLTISSLQPEDFATYYCQQHYTTPPTFGQGTKVEIKRTVAAPSVFIFPP
SDEQLKSGTASVVCLLNNFYPREAKVQWKVDNALQSGNSQESVTEQDSKDSTYSLSSTLT
LSKADYEKHKVYACEVTHQGLSSPVTKSFNRGEC

>Anti-HER2 Heavy chain (1 and 2)
EVQLVESGGGLVQPGGSLRLSCAASGFNIKDTYIHWVRQAPGKGLEWVARIYPTNGYTRY
ADSVKGRFTISADTSKNTAYLQMNSLRAEDTAVYYCSRWGGDGFYAMDYWGQGTLVTVSS
ASTKGPSVFPLAPSSKSTSGGTAALGCLVKDYFPEPVTVSWNSGALTSGVHTFPAVLQSS
GLYSLSSVVTVPSSSLGTQTYICNVNHKPSNTKVDKKVEPKSCDKTHTCPPCPAPELLGG
PSVFLFPPKPKDTLMISRTPEVTCVVVDVSHEDPEVKFNWYVDGVEVHNAKTKPREEQYN
STYRVVSVLTVLHQDWLNGKEYKCKVSNKALPAPIEKTISKAKGQPREPQVYTLPPSREE
MTKNQVSLTCLVKGFYPSDIAVEWESNGQPENNYKTTPPVLDSDGSFFLYSKLTVDKSRW
QQGNVFSCSVMHEALHNHYTQKSLSLSPGK

Summarize the risks of W35 oxidation with respect to the 4 following risk categories: biological activity, PK/PD, immunogenicity and safety.

Short answer:

For this IgG1, isolated oxidation of heavy-chain W35 (Kabat, in VH framework 2, “DTYIHWV…”) is generally a low-to-moderate risk liability. Main concern is local Fab destabilization and possible impact on antigen affinity; Fc-mediated PK and classic safety signals are much less directly affected.

Long answer:

1. Biological activity (target binding, effector function)

Location / role

  • W35 is in the VH framework 2, adjacent to CDR-H1/2, not in Fc.

  • It contributes to the hydrophobic core that positions the CDR loops and the VH–VL interface rather than directly contacting HER2.

Effect of oxidation

  • Tryptophan oxidation → hydroxy-Trp, N-formylkynurenine, kynurenine:

    • Increases polarity in a mostly hydrophobic environment.

    • Can slightly destabilize the VH framework and alter CDR geometry.

  • At low levels (a few percent of molecules oxidized at W35):

    • Expected effect on HER2 binding affinity is typically small.

    • ADCC/CDC are largely unchanged if antigen binding and Fc remain intact.

  • At high levels or if oxidation correlates with partial unfolding:

    • Could reduce affinity and cell-based potency by mis-positioning CDR-H1/H2 or VH–VL.

Risk assessment for biological activity

  • Intrinsic risk: Low to moderate.

  • More likely to produce a gradual potency drift than a binary loss of function.

  • Risk is best quantified by forced-oxidation + binding / cell-based potency comparison and setting a control limit on oxidized variants if needed.

2. PK/PD

Direct mechanisms

  • W35 is in the Fab variable domain, not in CH2/CH3 where:

    • FcRn binding sites reside.

    • Classical Fc oxidation hot spots (e.g., M252/M428, W313) live.

  • Direct effects on FcRn-mediated recycling, nonspecific Fc receptor interactions, or charge-based clearance are expected to be minimal.

Indirect mechanisms

  • If W35 oxidation significantly destabilizes the Fab, possible consequences:

    • Slightly increased fragmentation or aggregation.

    • Enhanced clearance of aggregates or partially unfolded species.

  • At realistic in-product levels (e.g., ≤5–10% oxidized at that site), these effects on systemic PK are usually smalland may be within normal batch-to-batch variability.

Risk assessment for PK/PD

  • Direct PK risk: Low.

  • Indirect PK/PD risk via instability/aggregation: Low to at most moderate, contingent on empirical data.

  • Focus PK risk control more on Fc and high-level aggregation than on this single Fab Trp site.

3. Immunogenicity

Potential immunogenic mechanisms

  1. Neo-epitope formation

    • Oxidized Trp can slightly change local conformation and side-chain chemistry.

    • The region is in VH framework 2, partly solvent-exposed.

    • This can, in principle, alter MHC-II peptide repertoire and T-cell recognition.

    • However, clinical evidence that single-site Trp oxidation in Fab alone drives ADA is limited.

  2. Aggregation-driven immunogenicity

    • If W35 oxidation destabilizes VH and increases:

      • Reversible or irreversible aggregates

      • Subvisible particles

    • These species are more relevant immunogenicity drivers than the chemical modification itself.

Risk assessment for immunogenicity

  • Intrinsic chemical neo-epitope risk at W35: Theoretical but low at typical levels.

  • Aggregation-mediated immunogenicity:

    • Depends on whether W35 oxidation correlates with increased aggregates under real storage conditions.

    • Usually managed by controlling total aggregates and oxidized variants.

Overall: Low immunogenicity risk, provided:

  • W35 oxidation is kept in a controlled range.

  • Aggregates and subvisible particles remain within standard limits.

  • Clinical ADA monitoring does not show a correlation with high-oxidation lots.

4. Safety (including off-target effects)

Direct safety concerns

  • W35 is not in Fc; no direct link to:

    • Complement overactivation.

    • FcγR hyper-engagement.

  • Oxidation at this site is unlikely to introduce a strong new off-target receptor interaction.

Indirect safety concerns

  • Same mechanisms as immunogenicity:

    • More aggregates or fragments → potential for infusion reactions or ADA-mediated effects.

    • Loss of binding specificity (if Fab is destabilized) could, in extreme cases, slightly increase nonspecific binding, but this typically appears only under harsh stress and very high oxidation.

Risk assessment for safety

  • Direct safety risk: Very low.

  • Indirect safety risk:

    • Tied to aggregation and immunogenicity.

    • Expected to be low if product-quality attributes (oxidation levels, aggregates, particles) meet standard biologics limits and no unusual clinical signals are seen.

Practical takeaways

  • W35 oxidation is a secondary Fab liability, not a primary one like Fc Met oxidation.

  • Key control strategy:

    • Characterize W35 oxidation in forced-degradation and long-term stability.

    • Correlate with:

      • HER2 binding / cell-based potency.

      • Aggregate and fragment levels.

      • Any ADA or safety signals, if clinical data exist.

  • If no strong correlations emerge, W35 oxidation can typically be treated as a monitored but not critical quality attribute with relatively broad acceptance limits compared with Fc oxidation hot spots.

Simon Letarte
Next

First Post