CFU Testing for Probiotics: What Your COA Really Proves

A CFU figure on a certificate of analysis reads like a fixed property of the powder, but it is not. It is the output of a chain of decisions: where the sample came from, how it was rehydrated and homogenised, which diluent was used and how long cells sat in it, which counting method was applied, and how the result was calculated. Change one link and the same lot can report several times higher or lower, occasionally by an order of magnitude. For a brand owner that spread is not academic: it decides whether the label claim is defensible, whether the release specification is achievable, and how much extra potency you must buy to remain compliant at expiry.

The three methods used most often do not measure the same thing. A plate count enumerates cells able to form visible colonies under one defined set of medium, temperature, atmosphere and incubation time; stressed or viable-but-non-culturable cells stay invisible, and a clump of cells registers as a single colony. Flow cytometry with membrane-integrity dyes counts cells with an intact envelope, whether or not they would grow, and typically returns a higher figure. qPCR amplifies a DNA target and reports gene copies rather than colony-forming units; without a viability dye it cannot separate live cells from dead ones. Each method is legitimate, and none is the true count.

A CFU number is not a property of the powder alone. It is a property of the powder plus the method used to count it.

The second ambiguity is the reference point. Some markets expect the number on the label to hold until the end of shelf life; others accept an at-time-of-manufacture statement when the wording says so explicitly. The two conventions produce very different overage requirements, and mixing them is where budgets quietly break. Before agreeing a specification, confirm which convention your label uses, which one your contract's release limit assumes, and whether the wording on pack matches both.

A one-order-of-magnitude disagreement between a supplier's COA and a marketplace or retailer test is expensive in three places at once. Potency-bearing material is priced per unit of activity, so overage scales almost linearly with the count you believe you need. A lot that tests low downstream can trigger rejection, rework or refunds on product already packed. And repeated discrepancies damage the account even when each individual dispute is settled. The remedy is not a stricter number but a defined method, a defined sampling plan and a defined tie-breaker.

Sampling and sample preparation are where most disagreements are born. Probiotic powders are rarely homogeneous: fine fractions stratify during blending and transport, hygroscopic powders clump, and static pushes particles to container walls. A single scoop from the top of a drum is not a lot sample. Composite increments from several containers, a defined sample mass and a documented quartering or riffling step are the minimum. Rehydration medium, its temperature and the resuscitation time before plating all affect recovery, and so does the interval between dilution and plating; cells left too long in an unbuffered diluent lose viability that the plate count then reports as a manufacturing loss. Homogenisation must be strong enough to release cells from the matrix and gentle enough not to injure them.

Multi-strain blends are harder still, because a general medium counts everything that grows on it. A total viable count on a broad-spectrum medium says little about whether each declared strain is present at its declared share. Enumeration therefore has to be strain-directed: validated selective media keyed to a distinguishing phenotype, or species- and strain-specific qPCR assays, run alongside a total count as a reconciliation check. Whatever scheme is chosen needs documented recovery data for each target strain in the presence of the others, plus a stated tolerance for how the sum of individual counts should compare with the total. Ask whether the COA reports per-strain counts or only a sum.

The dosage form changes the recovery problem again, so each format needs its own validated pre-treatment rather than a borrowed one. Hard capsules are comparatively simple, but the shell or beadlet coating must be fully dispersed before dilution. Softgels suspend cells in oil, which has to be released and the cells partitioned into an aqueous phase without osmotic or surfactant damage. Gummies combine low water activity, sugar alcohols and acidulants, and need controlled dissolution and pH adjustment before plating; they also lose viability during deposition, so their overage logic differs from a powder's. Stick sachets are dry but hygroscopic, and clumping during storage can make sampling error larger than method error. Water activity and packaging barrier belong in the specification, not only in the marketing copy.

CFU is only half of the credibility question; identity is the other half. 16S rRNA sequencing resolves genus and often species but rarely strain, so it cannot confirm that the organism in your capsule is the licensed strain named on the label. Whole-genome sequencing with genome-wide comparison, or established fingerprinting methods such as RAPD and rep-PCR, can go further, though fingerprints travel poorly between laboratories. Strain-level confirmation is warranted when qualifying a new raw material source, when a license names a specific strain, and whenever a label carries a strain designation. That designation must be identical in the formula, the label and the COA. A usable COA also states lot number, method reference, sample preparation, incubation conditions, result units and the analyst and date; retention samples of at least two sealed units per lot, stored as labelled, should remain available past expiry.

Contract language turns all of this into something enforceable. Set the release specification and the shelf-life specification separately: release sits above the label claim, expiry equals it, and the gap is derived from measured stability loss plus method and sampling variability rather than a familiar rule of thumb. The stability package behind that gap should cover several production lots, real-time storage at the labelled condition as primary evidence, accelerated data as support only, and the actual container-closure system. Keep the right to retest any lot at an accredited third-party laboratory using the agreed method, with a defined split-sample procedure, a named tie-breaker lab and clear allocation of costs and remedies if the retest falls short. Then ask your manufacturer, in writing, which method they use, how they sample, whether counts are per strain, and what their stability data actually shows.