FDA Warning Letter 320-26-40 to Cohance Lifesciences did not turn on a single residue number. It turned on whether the firm's verification programme could have found the carryover at all — and whether its worst-case basis was ever revisited.
What happened
On 30 January 2026 the US Food and Drug Administration issued Warning Letter 320-26-40 (MARCS-CMS 718812) to Cohance Lifesciences Limited, formerly Suven Pharmaceuticals, covering the finished dosage form facility (FDF Unit-I) at Plot No. A-19/C, A-23A, A-23B, Road No. 18, IDA Nacharam, Uppal Mandal, Medchal-Malkajgiri, Hyderabad (FEI 3008768274). The inspection ran from 4 to 12 August 2025 and closed with a Form FDA 483 carrying six observations. The site was classified Official Action Indicated, and the warning letter followed from that classification. The company disclosed receipt of the letter on 4 February 2026 and stated that US revenues from the facility contributed less than 2% of consolidated revenues in FY2025.
The letter was issued by CDER and states that drug products are adulterated within the meaning of section 501(a)(2)(B) of the FD&C Act because methods, facilities or controls for manufacturing do not conform to CGMP. FDA reviewed the firm's 2 September 2025 response and found it inadequate.
Two of the cited violations sit squarely in cleaning validation territory.
Under 21 CFR 211.67(a), the Warning Letter identifies visible stains inside a duct serving a piece of equipment (PD/(b)(4)-01) and visible residues on a capsule filling machine (PD/CFM-03) that had been documented as clean and had already undergone changeover cleaning. Analytical testing of those stains and residues confirmed the presence of multiple active pharmaceutical ingredients at values exceeding the firm's allowable limit. FDA also noted that the response did not address reserve sample testing of drug products released to the US market and still within expiry.
Under 21 CFR 211.67(b), the Warning Letter identifies a 2018 cleaning validation study that failed to use the identified worst-case drug product, failure to re-evaluate the cleaning validation programme when new drug products were introduced onto shared manufacturing equipment, and the absence of routine cleaning verification. The decisive sentence is short: your cleaning verification lacks the ability to detect cross-contamination from carryover.
Why it matters
The interesting failure here is not that residues were found. It is that the residue was found by an investigator's eye, not by the firm's own system — on equipment that the documentation called clean.
That inverts the usual order of defence. Visual inspection and analytical verification are supposed to function as two independent opportunities to catch carryover. In this case both were nominally in place and both failed to catch what an inspector could see unaided. FDA's response focuses accordingly: not on tightening a number, but on whether the verification programme has any capability to detect the risk it is supposed to control.
A verification method that cannot detect the relevant residue is not a weak control; it is an absent one that produces a compliant-looking record. That distinction is what makes this letter worth reading outside the specific firm.
What is particularly interesting
Three threads stand out.
First, the worst-case question. The Warning Letter states that the 2018 study did not use the identified worst-case product. That is a different kind of finding from "the limit was too high": the study design itself was disconnected from the firm's own hazard analysis. A cleaning validation built on a non-worst-case product answers a question nobody asked.
Second, the lifecycle failure. FDA explicitly identifies the absence of re-evaluation when new products were introduced onto shared equipment. Cleaning validation is not a documentation asset created once and filed. Adding a highly potent, poorly soluble, or difficult-to-clean product changes the risk profile of every shared equipment train it touches. If the programme does not react, the existing validation is describing a manufacturing operation that no longer exists.
Third, the analytical angle. Among the requested remediations, FDA asks the firm to demonstrate the adequacy of its analytical methodology to identify residual carryover, and to investigate unknown (unidentified) peaks detected in reserve samples. Unknown peaks are the analyst's early warning that a method is seeing something it was not designed for. That request converts method capability from a background assumption into a documented, reviewable claim.
The letter also lists the worst-case dimensions it expects to see addressed: drugs with higher toxicities, drugs with higher potencies, drugs of lower solubility in their cleaning solvents, drugs with characteristics that make them difficult to clean, swabbing locations for areas most difficult to clean, and maximum hold times before cleaning. Notably, hold time before cleaning appears alongside potency and solubility — a reminder that worst case in a shared facility is a multi-dimensional problem, not a single product pick.
FDA further requests a comprehensive independent retrospective assessment of cleaning effectiveness, covering the identity of residues, other equipment possibly improperly cleaned, and whether cross-contaminated products may have been released; a CAPA plan grounded in that assessment; and an independent consultant review of CAPA effectiveness.
How such a gap can realistically develop
A programme like this rarely collapses in one step. The more common path is incremental. A cleaning validation is performed when the equipment train is small and the product range narrow. The product portfolio expands; each addition is assessed as an isolated technical change rather than as a change to the cleaning risk profile. The original worst-case justification is never revisited because nothing triggers it. Meanwhile the verification method is carried forward unchanged, and its limits of detection are assumed rather than demonstrated.
Over time, the documentation becomes more confident than the underlying data. When a visibly dirty surface is signed off as clean, the failure is not that someone falsified a record. It is that a second check — analytical verification — had no realistic chance of disagreeing. Two controls that cannot fail independently are one control.
Questions for our programme
These are the questions this letter invites us to ask of our own sites, without assuming the answer is comfortable.
- Is our worst-case product selection documented, still current, and formally re-opened whenever a new product is introduced onto shared equipment?
- Can we demonstrate, with data, the limit of detection of our verification methods for the specific residues they are expected to find — not just that the method passes on a clean swab?
- What does our change control do when a new product is more potent, less soluble in the cleaning solvent, or harder to clean than the current worst case?
- Who owns revalidation triggers, and how do we prove that a portfolio expansion has been assessed against the cleaning validation lifecycle?
- Do we investigate unknown or unidentified peaks in cleaning and stability samples as method-capability signals, or close them as noise?
- How do we test whether visual inspection and analytical verification can actually disagree — and is that ever exercised deliberately?
Wider context
This letter is not an outlier. The pattern — worst-case selection that was never updated, verification methods whose detection capability is assumed, and cleaning validation treated as a fixed document rather than a lifecycle — recurs across recent FDA warning letters to shared-equipment oral solid dose facilities. The regulatory direction of travel is consistent: health-based exposure limits, science- and risk-based worst-case justification, hold times before cleaning, and demonstrated capability of analytical methods.
Our practical reading is measured. Nothing in this letter sets a number that applies to every facility, and nothing here implies that any single verification technique is sufficient. The transferable point is the one FDA made explicit: a cleaning verification programme exists to detect carryover, and its adequacy is a question of detection capability, not paperwork.
Sources
- FDA Warning Letter 320-26-40, Cohance Lifesciences Limited, MARCS-CMS 718812, 30 January 2026: https://www.fda.gov/inspections-compliance-enforcement-and-criminal-investigations/warning-letters/cohance-lifesciences-limited-718812-01302026
- Cohance Lifesciences Limited, SEBI Regulation 30 disclosure (Form A) on the USFDA warning letter, 6 February 2026: https://www.cohance.com/wp-content/uploads/2026/02/Cohance_SE_intimation_USFDA_warning_letter_06022026.pdf
- BSE India, company disclosure on receipt of the USFDA warning letter, February 2026
- MarketScreener, reproduction of Warning Letter 718812 text, 7 February 2026
- The Hindu, "U.S. FDA issues warning letter to Cohance's FDF facility in Hyderabad", 4 February 2026
Validation takeaway
A cleaning verification programme is only as good as its ability to detect the carryover it is meant to control. FDA's letter to Cohance turns on two related gaps: a worst-case selection that was never revisited as products were added to shared equipment, and verification that could not detect cross-contamination from carryover. Method capability and worst-case currency are not administrative details — they are the evidence that the control exists at all.
Questions worth asking in our own programme
- Is our worst-case product selection documented, current, and formally re-opened whenever a new product enters a shared equipment train?
- Can we demonstrate the limit of detection of each verification method for the specific residues it is expected to find?
- What does change control do when a new product is more potent, less soluble in the cleaning solvent, or harder to clean than the current worst case?
- Who owns revalidation triggers, and how is portfolio expansion assessed against the cleaning validation lifecycle?
- Do we treat unknown or unidentified peaks in cleaning and stability samples as method-capability signals rather than closing them as noise?
- Have we ever demonstrated that visual inspection and analytical verification can disagree — and exercised that deliberately?
