Naming and counting unknowns.
+ The questionHow do instruments identify compounds no one listed in advance?
Separate first, then identify
In FDA staff's example of Chapter 12, a first extraction cycle can leave 10 mg, a mass that carries no names. The dried residue of the running device's cushion extract could be one compound or dozens, and a balance cannot tell which. An instrument pointed at the whole extract would meet the same problem in another form, because a mixture of compounds gives a mixture of signals that overlap. The analysis therefore starts by pulling the mixture apart, and only then asks what each part is.
Chromatography pulls a mixture apart by partitioning, the sharing of Chapter 4. A moving gas or liquid carries the extract through a narrow column whose wall or packing holds a stationary layer, and each compound shares itself between the moving phase and that layer. A compound that dissolves better in the layer spends more of its time there and travels more slowly, so compounds that entered together leave the column one after another. A detector at the outlet records a signal against time, the chromatogram, in which each compound appears as a peak. The time a compound takes to emerge is its retention time, and it depends on the compound, the column and the conditions. A peak's area grows with the amount of its compound.
A peak alone names nothing, since different compounds can emerge at the same time. The column therefore feeds a mass spectrometer, which gives each emerging compound an electric charge, sorts the charged pieces by mass and counts how many arrive at each mass. The pattern of masses is the compound's mass spectrum, and it follows from the compound's structure. The chromatogram thus says when each compound arrived and how much of it there was, and the spectrum gives the evidence of what it is.
Each instrument sees its own kind of compound
A gas can carry only what evaporates. Gas chromatography heats its column and moves the extract through it in a stream of gas, so it reaches compounds small enough to evaporate at its temperatures, the volatile and semi-volatile ones. Coupled to a mass spectrometer, gas chromatography becomes GC-MS, and FDA's 2024 draft guidance on chemical analysis assigns these compounds to GC-MS in polar, semi-polar and non-polar extracts alike (Chapter 12). The most volatile compounds can be missed when a liquid extract is injected, so headspace GC-MS samples the gas above a sample instead. The draft names headspace GC-MS as a complement for polar solvents and for test articles with no solvent at all, and its examples are the volatile solvents of adhesives and silicone cyclic oligomers, short silicone chains closed into rings (Chapter 3).
A liquid carries what does not evaporate. Liquid chromatography moves the extract through its column in a liquid, so it reaches larger, non-volatile compounds and some semi-volatile ones. For these the draft names LC-MS with electrospray as the primary method, run in both its positive and its negative mode. Electrospray charges molecules by adding or removing a charge while leaving them mostly whole, and both modes matter, because a compound that takes a charge poorly in one mode may show only in the other. Elements such as metals need a third approach. Inductively coupled plasma mass spectrometry, ICP-MS, heats a polar extract in a plasma, a hot ionized gas that breaks molecules into charged atoms, and the draft assigns elements to ICP-MS or to its optical counterpart, ICP-OES.
No technique sees everything, which is why each extract goes to more than one. In a 2025 bench study, Yun and colleagues, chemists at FDA, ran 106 polymer additives through GC-MS and LC-MS, and FDA alone funded the work. Of 92 additives analyzed by GC-MS, 17 fell below detection or were not detected. In LC-MS, 12 of 57 gave no usable signal in positive mode and 19 of 28 in negative mode, as the paper's table counts them. The paper's text gives the negative-mode count as 19 of 27. A compound that no technique detects never reaches a toxicologist. Nelson Labs, a testing laboratory, called such a miss an error of omission in workshop slides in 2023.
A spectrum names a compound at a level of confidence
A laboratory names a compound by matching its spectrum against a library of spectra recorded from known compounds. In GC-MS, electron impact charges the molecules and breaks each into fragments, and the library matches the pattern of fragment masses. Yun and colleagues used electron impact at 70 eV. The libraries are large, and two that Nelson Labs' slides cite hold more than 300,000 and more than 840,000 GC-MS spectra, so a match yields only a candidate, which a mass spectrometrist checks before it becomes a name.
A match score measures how alike two spectra are. An extremely large library holds many similar spectra, and Nelson Labs' slides show a top hit with a high match score that was a wrong identification. The slides strongly recommend careful review by a mass spectrometrist before a match is accepted as a name.
How far the evidence goes sets the identification level, and FDA's draft names four. A compound is confirmed when its structure has been verified with an authentic reference standard of that compound, confident when the evidence allows a single candidate structure, tentative when one or more candidate structures can be assigned, and unknown when none can. A name that gives only a class of compound counts as unknown in the draft. The draft recommends confident or confirmed identities for the toxicological assessment. Nelson Labs grades more finely. Its 2023 slides set out a ladder of five rungs, with a partial identification between unknown and tentative, and cite a chapter of the US Pharmacopeia for the ladder.
Counting without a standard for every compound
A confirmed compound can be counted against its own standard: the laboratory runs known amounts of it and reads the extract's peak against them. The other compounds have no standard of their own. For them the laboratory adds a known amount of an internal standard, one chosen compound that serves as a surrogate, and counts each peak as if its compound gave the same signal per microgram as the internal standard. For equal amounts, the ratio of a compound's signal to the internal standard's is the compound's relative response factor, and counting by the internal standard assumes that ratio is 1.
Compounds do not respond alike, so the assumption miscounts them. In GC-MS, Yun and colleagues measured the additives against one internal standard at 5, 10 and 20 µg/mL, and the weakest-responding additive gave 0.014 times the internal standard's signal. Counted by the internal standard, that additive would be reported at 0.014 of its true amount, so its true amount is about 71 times the reported one, while the strongest responder would be reported at about 2.6 times its true amount. LC-MS spread the responses wider. No molecular property tracked the response closely, and the authors attribute the lack of a pattern to how differently molecules take up a charge. In GC-MS the response fell only weakly to moderately as molecular weight and boiling point rose, and in LC-MS it showed no significant trend.
The laboratory therefore allows for the spread instead of predicting each response. It measures the response factors of the many compounds it holds standards for and expresses their spread as the relative standard deviation, the standard deviation as a fraction of the average. From that spread it computes the analytical uncertainty factor, the average response divided by the response of a compound one standard deviation below the average. A wider spread gives a larger factor. Once the spread reaches the average, a compound one standard deviation below the average would give no signal at all, and ISO's 2022 amendment on the factor states that the formula then gives no usable result. The analytical factor is not the toxicologist's uncertainty factor of Chapter 14, which divides a dose. Chapter 17 divides the reporting threshold by the analytical factor.
The factor has no agreed value. ISO's amendment notes that a factor of 1 can be justified where response is known to vary little, and that a factor of 2 "has been proposed" for GC screening. For GC-MS, FDA's draft allows a default "as low as 4" without further justification. For headspace GC, Oktem and colleagues, scientists at FDA, measured the response factors of polymer additives, and the initial data they presented in a 2021 poster gave factors of 6 to 10. For LC-MS, neither FDA's draft nor the amendment sets a value, and laboratories derive their own from databases of their own response factors.
- The rule. The factor is 1 divided by the result of 1 minus the relative standard deviation, the form that FDA's draft states.
- GC-MS. Nelson Labs reported a spread of 50 % across 84 semi-volatile compounds at FDA's 2024 meetings on ASCA, the accreditation scheme of Chapter 25. 1 minus 0.50 is 0.50, and 1 divided by 0.50 is 2.0.
- LC-MS with APCI, atmospheric pressure chemical ionization, another way of charging molecules: 63 compounds, spread 74 %. 1 minus 0.74 is 0.26, and 1 divided by 0.26 is about 3.8.
- LC-MS with electrospray: 50 compounds, spread 115 %. 1 minus 1.15 is less than zero, so the rule gives no factor. Nelson Labs reported a factor of 10, and its slides do not say how.
- The result. One laboratory's three methods give 2.0, about 3.8 and no value by the rule: the factor rises with the spread until the rule fails.
- Assumptions. The spreads come from the laboratory's own databases, and Nelson Labs' slides stress that these values are specific to each laboratory. The rule presumes a roughly normal spread of response factors, ISO's amendment notes.
What the report shows, and what it leaves open
A laboratory's report carries the evidence behind every name and amount. FDA's draft asks for complete reports: the method and its justification, the extraction conditions and results, the non-volatile residue, the instrument system, the calibration, the chromatograms, and the identities and amounts of the extractables. The non-volatile residue of Chapter 12 checks the sum, because it estimates by weight the non-volatile and some of the semi-volatile extractables, so identified amounts that fall far short of it can point to compounds missed or miscounted. The draft also asks that every extractable above a reporting threshold be quantified and reported, and Chapter 17 sets that threshold.
The running device's cushion shows what such a report can leave open. Its extracts give three kinds of result, set out below without amounts. The results are this book's illustration, and no maker's record supports them. Only the first result goes to toxicology with a name, a registry number and an amount counted against its own standard. Chapter 14 sets that compound's tolerable intake.
| Result in the cushion's extracts | Identification level | How its amount is counted |
|---|---|---|
| D4, octamethylcyclotetrasiloxane, a silicone ring (CAS 556-67-2), by headspace GC-MS of the water extract | confirmed | against a D4 standard |
| a peak in the hexane extract whose spectrum fits two library entries, by GC-MS | tentative | by the internal standard |
| a peak in the isopropanol extract that matches no library spectrum, by LC-MS | unknown | by the internal standard |
The other two results carry both of this chapter's uncertainties, one in the name and one in the amount. The tentative compound has two candidate structures, and the unknown peak has none, so neither can be looked up as one compound. Each amount comes from the internal standard, which may understate it many times over. Chapter 16 takes up what can be done with each of the two.
Before an extract goes to a laboratory, agree which techniques each solvent's extract will get, from GC-MS, headspace GC-MS and LC-MS in both modes to ICP-MS, and how each will be calibrated. Ask for every peak above the reporting threshold with its identification level and the evidence for it, the standard or internal standard behind its amount, and the analytical uncertainty factor with the database it comes from. Compare the identified total with the non-volatile residue, and ask what a second method could add for each tentative or unknown peak.
- Chromatography separates an extract's compounds by how each shares itself between a moving phase and a stationary layer, and a mass spectrometer then records each compound's spectrum.
- GC-MS reaches volatile and semi-volatile compounds, headspace GC-MS the most volatile, LC-MS larger non-volatile ones and ICP-MS elements, and no technique detects every compound.
- A library match yields a candidate, and FDA's draft grades each identification as confirmed, confident, tentative or unknown, with a class-only name counted as unknown.
- A compound without its own standard is counted by an internal standard, which can miscount it many times over, and the analytical uncertainty factor that allows for the spread of responses has no agreed value.
Let's Talk