Biologics degrade under cleaning, ADCs split one residue into antibody, linker, payload and conjugate, and peptide APIs combine high potency with methods already under strain. A synthesis of the local evidence on why toxicology, cleaning and detectability now move together.
Cleaning validation was designed around a small-molecule assumption: one parent API, one toxicity number, one measurable residue. Biologics, antibody-drug conjugates and peptide APIs each break a different link in that chain — and they break it at the same time.
Why new modalities break old assumptions
The conventional chain is settled. A permitted daily exposure is derived from toxicological data, converted into a maximum allowable carry-over value using the maximum daily dose and minimum batch size of the subsequently manufactured product, distributed across shared product-contact surface area, corrected for recovery, and enforced by a validated method. ASTM E3219-25, the current edition of the HBEL guide (active, last updated 10 June 2025, DOI 10.1520/E3219-25), sets out the derivation sequence in its publicly visible section 4.5: hazard characterisation, critical effect with dose-response, point of departure, PoD-specific adjustment factors, and justification of the final value. Its scope clause 1.3 covers APIs including biologics, intermediates, cleaning agents and excipients.
That chain carries four implicit assumptions: the residue on the surface is the molecule that was dosed; its toxicity is known; a method exists whose limit of quantification sits below the limit; and the limit sits above the limit of visual detection. New modalities break these assumptions in three different places at once.
The regulatory baseline moved with them. PIC/S PI 006-4, published 30 July 2026 and in force from 1 October 2026, gives cleaning validation its own chapter — Chapter 7 — with section 7.3.1 making the extent of a cleaning programme dependent on a toxicological risk assessment, for example on HBEL values.
The degradation question (biologics)
For a biologic, the residue that survives cleaning is not the drug substance. Alkaline CIP at 0.5 to 1 N NaOH and 70 to 80 degrees Celsius denatures protein, so a product-specific immunoassay loses specificity and TOC becomes the primary residue indicator.
Cohen et al. (J Pharm Sci 2025;114(2):866-877; DOI 10.1016/j.xphs.2024.10.041) put that on a measured footing. Using a bench-scale cleaning method they generated degradants from one half-life-extended (HLE) BiTE molecule and one monoclonal antibody (mAb-1), characterised them alone and after spiking 650 micrograms of degradant into mAb-1 in same-product and cross-product backgrounds, and ran an orthogonal panel: endotoxin, size exclusion chromatography, light obscuration by HIAC and micro-fluidic imaging. Immunogenic potential was assessed with a cell-line immune activation assay and a human PBMC assay measuring T-cell activation, proliferation and cytokine release at 20 hours and 7 days.
The authors report that the cleaning procedure caused significant degradation, so the molecule was no longer intact or active, and that 650 micrograms of degradant carryover — alone or spiked into the same or a cross product — did not increase immunogenicity risk in the cell-based assays. Their conclusion: a PDE of 650 micrograms per dose for cleaning-induced degradant carryover presents no immunogenicity risk to patients. The value is not new: it traces to clinical safety experience with gelatin as a reference immunogen.
Two design features matter for programme work. The PBMC readout extends to 7 days, so it probes the late adaptive phase rather than an acute cytokine spike, and the degradant was tested in the presence of the subsequently manufactured product — the configuration that actually occurs at a multiproduct changeover.
The consequence is direct: for biologics the relevant PDE is the degradant PDE, not the active-molecule PDE. A limit set on intact protein can fall below both the capability of the cleaning process and the quantification limit of the method meant to enforce it, making it undemonstrable in both directions; a degradant value roughly two orders of magnitude higher converts an unverifiable swab limit into something the process can prove. The limitations are clear too: the work is entirely in vitro, covers one HLE BiTE and one mAb, and relies on a bench-scale surrogate whose equivalence to production CIP is argued rather than demonstrated.
The payload question (ADCs)
An ADC is not one residue. It comprises the antibody, the chemical linker, the free payload, the intact conjugate and the degradation products of each — and under the same cleaning chemistry those components behave differently, so a shared-equipment strategy developed for the antibody may be largely irrelevant to a lipophilic auristatin or PBD payload. HPAPI payloads used in ADCs sit in Category 4 of the SafeBridge-derived occupational classification, the most hazardous category, and Bioprocess International reports that HPAPI facilities require MACO below 10 ppm.
The EMA HBEL guideline is explicit that for a therapeutic protein conjugated to a small-molecule pharmacophore, the toxicity of the conjugate must be considered: a PDE derived only for the antibody, or only for the free payload, does not describe the residue risk.
Linker lability compounds the problem. Peer-reviewed LC-MS work on ADC bioanalysis confirms that cleavable linkers release payload and that cleavage during sample processing can artificially inflate measured free-payload values: a swab extract is not a stable system, and the analyte quantified may not be the residue that was on the surface. Hold-time data for linker-labile products is a false-negative control, not a formality.
Analytical sensitivity, by contrast, is often the easier half of the problem: Samsung Biologics, publishing its own cleaning-method development, reports LOQ and LOD values well below typical ADC residual acceptance limits. The pitfalls independent guidance lists are recovery and limit-setting issues rather than detection issues: critical cleaning parameters, drug product degradation, recovery factors, air-liquid interface residues, TOC recovery and stainless-steel maintenance.
The peptide question
Peptide APIs sit between the two. Semaglutide and tirzepatide are high-potency molecules made on shared equipment, and low doses drive low permissible carryover — the situation health-based limits were designed for.
FDA Warning Letter 320-26-73 (MARCS-CMS 723330, dated 1 May 2026), after an inspection of Harbin Jixianglong Biotech Co., Ltd. in November 2025, describes the failure mode precisely. The firm's cleaning procedure required calculation of a maximum allowed carry-over limit based on API toxicity, yet the validation study for equipment used to manufacture peptide APIs, including semaglutide and tirzepatide, established no such limit, provided no justification for the swab sampling locations and contained no swab results. FDA found the firm's response inadequate because it lacked a systemic corrective and preventive action for the misalignment between procedure requirements and validation study.
The instructive part is that the gap is not a missing procedure: it is a procedure and a study that stopped agreeing, something a quality unit can detect before an inspector does by reading the SOP and the validation report side by side. The same inspection cited method validation gaps for assay and related substances by HPLC, aggregates by SEC, amino acid ratio and bacterial endotoxin. Residue methods for peptides must be validated for recovery and specificity, and the resolution weaknesses cited for related substances tend to reappear in cleaning support methods.
The detection question
The fourth break is detectability, and here the regulatory position has become structural.
PIC/S PI 006-4 requires, in section 7.13.2, that products whose maximum allowable carry-over limits fall below the limit of visual detection, safety factor included, be subject to cleaning verification after every change-over. Section 7.8.3 states that visual inspection will likely not be appropriate for lower-HBEL products or where visual thresholds are nearer to the required safety limit. Section 7.9.2 notes that swab sampling is generally more effective than rinse sampling and should be used where feasible, because rinse results are averaged and high-residue areas may not be identified. For potent payloads and for peptide APIs, those three provisions point in the same direction.
The demands land in the QC laboratory, where cleaning support methods compete with release testing for the same capacity. Moura et al. (DARU J Pharm Sci 33:20, 2025; DOI 10.1007/s40199-025-00566-x) built a structured cleaning-validation protocol for 17 frequently used pieces of laboratory glassware and stainless steel, using oxcarbazepine as worst-case API, separating compound-specific HPLC residue analysis — limit of quantification 0.005 micrograms per millilitre, limit of detection 0.002 — from detergent checks by conductivity and TOC. Swab recovery exceeded 70 percent after the first cleaning, rinse recovery exceeded 65 percent on stainless steel, and second-cleaning repeatability was poor across all conditions. The authors' stated limitation is the point for new modalities: one API, limited surface variety, and a need for more sensitive or multi-residue techniques such as LC-MS/MS for degradation products and excipients — exactly the analyte classes that biologics, ADCs and peptides bring into the laboratory.
What this means for programmes
Set the limit on what survives cleaning rather than on what was dosed. For biologics, document a degradant PDE distinct from the active-molecule PDE, and defend the equivalence between the study's cleaning conditions and production CIP.
Segment the ADC soil. Treat antibody, linker, free payload, conjugate and degradants as separate residue classes with separate acceptance rationales, and derive limits for the conjugate and free payload as distinct toxicological entities.
Treat recovery and sample stability as modality-specific. A recovery factor qualified on one material of construction and one analyte class does not transfer to a lipophilic payload or a labile linker.
Know what the method cannot see. Where a limit falls below the limit of visual detection, make verification after every change-over the documented routine rather than an exception.
Reopen the assumption chain on trigger. A new product, a higher-potency compound, a new linker-payload family or new toxicological data should recalculate the limit and re-open worst-case selection — not sit as narrative in a study written for an earlier product list.
Fazit
New modalities do not change the arithmetic of cleaning validation; they change the inputs. Biologics move the relevant PDE from the active molecule to its degradants, ADCs multiply one residue into several with different cleanability and different toxicological weight, and peptide APIs combine high potency with methods whose specificity and resolution are already under strain. In each case the binding constraint is less often analytical sensitivity and more often the correspondence between three documents: the limit, the method enforcing it, and the cleaning step that actually ran. ASTM E3219-25 and PIC/S PI 006-4 push in the same direction: the toxicological risk assessment determines the extent of the programme, and the evidence must cover what remains on the surface after the cleaning chemistry has acted.
Sources
- Cohen JR, et al. J Pharm Sci. 2025;114(2):866-877. DOI 10.1016/j.xphs.2024.10.041. PMID 39490658. https://pubmed.ncbi.nlm.nih.gov/39490658/
- STERIS Life Sciences webinar: Antibody Drug Conjugate (ADC) Cleaning Validation — Managing Complexity Across the Product Lifecycle, 9 July 2026. https://www.sterislifesciences.com/Events/2026/July/ADC-Cleaning-Validation
- ISPE — Cleaning Considerations for Antibody Drug Conjugate (ADC) Products. https://ispe.org/webinars/videos/cleaning-considerations-antibody-drug-conjugate-adc-products
- Regulated bioanalysis of antibody-drug conjugates using LC-MS (PMC12087923). https://pmc.ncbi.nlm.nih.gov/articles/PMC12087923/
- Samsung Biologics — A versatile analytical approach to cleaning validation for multiple linker-payload ADC platforms. https://samsungbiologics.com/media/science-technology/a-versatile-analytical-approach-to-cleaning-validation-for-multiple-linker-payload-adc-platforms
- Bioprocess International — Process Challenges of Antibody-Drug Conjugates. https://www.bioprocessintl.com/monoclonal-antibodies/process-challenges-of-antibody-drug-conjugates
- US FDA Warning Letter 320-26-73, Harbin Jixianglong Biotech Co., Ltd., MARCS-CMS 723330 (1 May 2026). https://www.fda.gov/inspections-compliance-enforcement-and-criminal-investigations/warning-letters/harbin-jixianglong-biotech-co-ltd-723330-05012026
- ASTM E3219-25, Standard Guide for Derivation of Health-Based Exposure Limits (HBELs), DOI 10.1520/E3219-25. https://store.astm.org/e3219-25.html
- EMA/CHMP/CVMP/SWP/169430/2012 — Guideline on setting health-based exposure limits for shared facilities. https://www.ema.europa.eu/en/documents/scientific-guideline/guideline-setting-health-based-exposure-limits-use-risk-identification-manufacture-different-medicinal-products-shared-facilities_en.pdf
- PIC/S PI 006-4, Recommendations on Qualification and Validation, in force 1 October 2026. https://picscheme.org/docview/11277
- Moura MJ, et al. Cleaning validation in pharmaceutical quality control laboratories. DARU J Pharm Sci. 2025;33:20. DOI 10.1007/s40199-025-00566-x. https://link.springer.com/article/10.1007/s40199-025-00566-x
Validation takeaway
New modalities do not change the arithmetic of cleaning validation; they change the inputs. For biologics the residue that survives cleaning is degraded, so the PDE that should drive the MAC calculation is the degradant PDE — supported by human-cell immunogenicity data at 650 micrograms per dose rather than the active-molecule PDE. For ADCs, one residue becomes several — antibody, linker, free payload, conjugate and degradants — each with different cleanability and different toxicological weight, and a swab extract of a linker-labile product is not a stable system. For peptide APIs, low doses drive low permissible carryover, and the recurring enforcement finding is not a missing procedure but a procedure and a validation study that no longer agree. In all three cases the binding constraint is less often analytical sensitivity and more often the correspondence between the limit, the method that enforces it and the cleaning step that actually ran.
Questions worth asking in our own programme
- For our biological products, is the residue target defined as the degraded fraction that survives cleaning — with an explicit degradant PDE distinct from the active-molecule PDE — or are limits still written against intact protein?
- Do we hold SEC, HIAC and MFI data for our cleaning-induced degradant characterisation, or does the whole residue argument rest on total protein by UV and TOC?
- If we rely on a degradant-based limit, can we defend worst-case equivalence between the bench-scale degradation study and the production CIP cycle of the equipment actually used?
- For each ADC, have we defined the residue target set explicitly — antibody, linker, free payload, intact conjugate and known degradants — with a separate acceptance rationale per component?
- Is our ADC limit derived for the conjugate and the free payload as distinct toxicological entities, or extrapolated from the antibody alone?
- What are our swab and rinse recovery factors on each material of construction for each analyte class, and do they cover the lipophilic payload and the labile linker rather than only the protein?
- Do we hold sample stability data for cleaning samples of linker-labile ADC products covering the real laboratory hold time, so that in-vial degradation cannot produce a false pass?
- For peptide APIs such as semaglutide and tirzepatide on shared equipment, can we point to the derived numeric carry-over limit, the swab location rationale and the actual swab results in the same validation report?
- Where a carry-over limit falls below the limit of visual detection, is cleaning verification after every change-over our documented routine rather than the exception?
- What trigger does our change control apply when a new modality, a higher-potency payload or a new linker-payload family enters a shared train — recalculation of the limit, revision of the cleaning matrix, re-opening of worst-case selection?
