Protein assay adulteration

Why did a milk test that measured nitrogen invite adding a plastic ingredient?

For more than a century, the standard way to state how much protein a food contains has been to measure its nitrogen and multiply. The Kjeldahl digestion, published in 1883, and the combustion method that later joined it both do this: they convert everything nitrogen-bearing in the sample into a form that can be counted, and report the total. Multiply by 6.25 — or 6.38 for dairy — and you have "crude protein".

The method is accurate, cheap, and was never secret. It is also what made adding melamine, an industrial compound used in resins and laminates, a profitable thing to do to watered-down milk. The puzzle is not that someone cheated. It is that a perfectly good measurement, applied honestly, created the opening.

Start with why the multiplication works at all. Protein is built from amino acids, every one of which carries nitrogen in its backbone, and across ordinary foods the nitrogen content of protein sits close enough to 16 percent that dividing by 0.16 — multiplying by 6.25 — gives a serviceable answer. This is a proxy: nitrogen is not protein, but in real food it tracks protein closely.

Now ask what holds that tracking in place. Nothing in chemistry guarantees it. It holds because, in the foods the factor was calibrated on, essentially all the nitrogen happens to be in protein. The relationship is an empirical regularity about a particular population of samples, not a law about nitrogen.

Here is the step that matters. Ask who can change that population. If the sample is simply presented by nature, the regularity holds and the proxy is fine. But if the sample is prepared by someone who knows how it will be judged, the population is no longer given — it is chosen. And the moment a proxy becomes a target that a supplier is paid against, it stops being a passive observation of the food and becomes a specification the food can be engineered toward.

So what would an adulterant have to look like to exploit it? It needs three properties, and they are worth listing because they explain the choice completely: high nitrogen content by mass, low cost, and no interference with anything else being tested. Melamine has all three. Its formula is C3H6N6, which works out to roughly 66 percent nitrogen by mass, against about 16 percent for protein. A gram of melamine therefore reads on the assay like something close to four grams of protein. Urea, used for the same purpose, is around 47 percent nitrogen. Nothing exotic is going on. The adulterant is simply the cheapest way to buy a high reading on the exact quantity being measured.

Notice what the cheater is not doing. They are not defeating the instrument, forging a certificate, or bribing an inspector. The laboratory reports the truth: this sample contains that much nitrogen. Every step downstream is also correct, given its assumption. The failure is entirely in the inference from nitrogen to protein, and that inference broke because the person supplying the sample had both a reason and an opportunity to break it.

The consequences were not confined to a false number. Melamine was implicated in pet food deaths in 2007, where it combined with cyanuric acid to form crystals that damaged kidneys, and then in the 2008 Chinese infant formula scandal, where official figures counted around three hundred thousand affected children and six infant deaths. A compound chosen purely for its analytical properties turned out to have toxicological ones too — there was never any reason to expect otherwise, since nobody had selected it for being safe.

Was this foreseeable? In outline, yes, and it had precedent: nitrogen-boosting adulteration of feed and food was known long before 2008. What is hard is not predicting that a proxy will be gamed but predicting which proxy and by what, because the adulterant is chosen after the fact by someone with more imagination about the specific test than the test's designers had about attackers.

The instinctive fix is to test for melamine, and regulators did add specific methods for it after 2008. That is necessary and insufficient, and the reason is structural: a targeted test defends against one compound, while the incentive that produced melamine remains pointed at whatever else satisfies the three properties. Defending a proxy by naming its known abuses is a race the defender enters one move behind.

The more durable responses go at the gap itself rather than at the substance. One is to measure the thing you actually care about: amino acid analysis, or a dye-binding or infrared method that responds to protein structure rather than to nitrogen atoms, so a nitrogen-rich non-protein contributes nothing. Another is non-targeted screening — take a broad fingerprint of the sample by spectroscopy and flag anything that does not look like milk, which catches unknown adulterants without needing to name them in advance. A third is to make the test unpredictable, since a proxy is only exploitable to the extent the supplier knows they will be judged on it.

One further clarification. The comfortable conclusion is that proxies are bad and should give way to direct measurement — but direct measurement is slower and dearer, which is precisely why the proxy was adopted. A proxy is a bargain struck against an assumed population of samples, and it needs revisiting whenever the suppliers of those samples gain a reason to change it.

• How non-targeted screening decides what counts as "not looking like milk" without a list of adulterants to search for.
• Where else a nitrogen-to-protein factor is quietly assumed, such as in animal feed labelling and food composition tables.

Every term the collection defines is gathered in the glossary.