The 70% problem: what chlorine makes in our drinking water that we cannot identify

Panyawut Tonanon

22 Jun 2026 • 6 min read

Your water utility publishes a Consumer Confidence Report every year. It tells you the four regulated trihalomethanes were below 80 µg/L, the five regulated haloacetic acids were below 60 µg/L, and that the water is "safe." Every year, the same reassuring numbers. The case is closed.

It is not closed. The total organic halogen on the same sample — the cumulative amount of chlorine, bromine, and iodine covalently bound to dissolved organic matter — is typically two to three times the sum of those nine regulated compounds. A 2025 Water Research paper, "Chasing disinfection byproducts through the pipes," tracked 66 DBPs through full-scale distribution systems. Of those 66, only a dozen or so are on any regulatory list. The rest appeared, decayed, and transformed over the days water spent between the treatment plant and your tap. We are not measuring them. We do not, in most cases, know what they are.

What TOX is and how we measure it

Total organic halogen is the field's bluntest instrument. The method is older than most of the readers of this blog: pass a water sample through activated carbon at low pH, the halogenated organic matter adsorbs, combust the carbon, the covalently bound halogens convert to halides, and titrate microcoulometrically. You now have a number, in micrograms per liter as chloride, that tells you the total halogen bound to organic matter. You do not know what molecules the halogen is in. You do not know how many of them there are. You have, in the most literal sense, an unknown number of unknown compounds, summed.

For a typical chlorinated surface water, TOX falls between 50 and 200 µg/L as Cl. The regulated 4 THMs and 5 HAAs account for 20 to 50 µg/L. The gap is the central analytical challenge of drinking-water chemistry, and it has been a known gap since the 1980s. The reason it has not been closed is not sensitivity — modern LC-HRMS detects halogenated features at low nanogram-per-liter levels — but combinatorial size. Every chlorine, bromine, or iodine substitution on every functional group in natural organic matter, times the additional nitrogen-containing products that form when chloramine is the disinfectant, gives a feature space of thousands. We are not under-instrumented. We are under-resolved.

The known DBP classes

The compounds we do know fall into four canonical classes, with a long tail underneath.

The trihalomethanes are the originals. Chloroform, bromodichloromethane, dibromochloromethane, bromoform — four molecules, all sharing the formula CHX₃. They form by a haloform-type reaction between free chlorine or bromine and activated ketone and β-dicarbonyl structures in natural organic matter. They are volatile, regulated, and by historical accident the most-measured class in drinking water.
The haloacetic acids are the second-tier regulated class, nine chlorinated and brominated species, with dichloroacetic and trichloroacetic acid typically the dominant pair. The five regulated HAAs are usually reported as a sum; the four unregulated mixed-halogen species are usually not reported at all.
The haloacetonitriles are the first nitrogenous class. Dichloroacetonitrile, dibromoacetonitrile, and their brominated and chlorinated cousins form preferentially under chloramine, which is more aggressive at incorporating nitrogen into the precursor NOM than free chlorine. They are not regulated in the United States, although they are in some other jurisdictions. They are, on a molar basis, more cytotoxic than the trihalomethanes they partially replaced when utilities switched disinfectants in the 1980s.
The haloacetamides, halonitromethanes, and other nitrogenous and oxygenated species round out the known inventory. The remaining 50% to 70% of the TOX is in compounds the analytical community has never named at all.

Across this known inventory, the toxicity ranking is one of the most reproduced results in the field. Plewa and Wagner, in a series of mammalian-cell cytotoxicity studies going back to the early 2000s, established that iodinated DBPs are orders of magnitude more cytotoxic than brominated DBPs, which are orders of magnitude more cytotoxic than chlorinated DBPs, on a molar basis. Iodoacetic acid sits at the top of their cytotoxicity index. When source-water iodide is present at even low microgram-per-liter levels, the disinfectant step is producing trace amounts of the most cytotoxic compounds in the entire DBP family. Those trace amounts are not in the regulated 11.

The chloramine pivot

In 1979, the US Environmental Protection Agency promulgated the Total Trihalomethane Rule, capping total THMs at 100 µg/L (US EPA, 1979). The rule triggered a wholesale shift by US utilities from free chlorine to monochloramine. Chloramine cuts regulated THMs by 50% to 80% in most distribution systems, and regulated HAAs by a similar margin.

What the rule did not do, and what its framers could not have known in 1979, is account for the corresponding increase in nitrogenous DBPs. Monochloramine is a milder electrophile than free chlorine, but it incorporates nitrogen into the reaction products. The shift moved the DBP profile along a different axis, away from the regulated THMs and HAAs, toward haloacetonitriles, haloacetamides, halonitromethanes, cyanogen halides, and N-nitrosamines. A 2025 distribution-system study in Water Research confirmed the asymmetry: total DBPs increased 32% over time under free chlorine, decreased 23% under chloramine, with the difference driven mostly by the regulated classes (Niehaves, E. et al., 2025). By the Plewa/Wagner index, a kilogram of haloacetonitriles is worth several kilograms of trihalomethanes. A 23% drop in total DBPs is unlikely to translate into a comparable drop in toxicological load.

NDMA is the chloramine-specific problem with its own regulatory response. The US EPA has a 10 ng/L health-based target for N-nitrosodimethylamine (US EPA, 2016); California has a 10 ng/L notification level. The compound is a probable human carcinogen (IARC Group 2A), and chloramine-driven nitrosamine formation is now the limiting factor in wastewater-reuse schemes that rely on chloramine for residual disinfection. The 1979 rule solved the 1979 problem. The 2026 problem looks different.

The 70%: Kendrick mass defect plots for halogenated screening

The central question for the analytical chemist is no longer "can we detect DBPs at relevant concentrations?" — modern HRMS instruments do that routinely. The question is how to find the 10,000 halogenated features in a single LC-HRMS injection when we don't know what any of them are.

The Kendrick mass defect plot is the field's effective visualization for this problem. For chlorinated compounds, the natural mass scale is H/Cl: every Cl substitution replaces a hydrogen with a chlorine, and the resulting mass defect has a characteristic value. Convert every feature in a full-scan run to the Kendrick mass on that scale and plot nominal mass against mass defect, and every member of a homologous chlorinated series sits on the same horizontal line. Brominated compounds, on a different scale, sit on a different line. Iodinated on yet another. The matrix, the non-halogenated organic background, the solvent noise — all of it scatters randomly across the plot. The halogenated features cluster into a few clear bands, easily distinguished from the background by eye. The M and M+2 isotope peaks of any chlorinated species sit exactly 1.997 Da apart, a pattern the trained eye picks up immediately.

This is the workflow that turns an LC-HRMS full-scan from "10,000 features I have to look at" into "200 halogenated features worth following up." The follow-up step uses the rest of the analytical stack this blog has covered. CCS — the third dimension of separation from the first post of this series — resolves isomers. The non-target screening workflow uses database lookup with m/z, retention time, and CCS within tolerance to assign high-confidence IDs to the 5–10% of features that match a database. The remainder — 90% or more of the halogenated features in a real chlorinated surface water — are flagged for MS/MS, structural elucidation, or prediction.

The bottleneck is the database. The chemical universe of halogenated DBPs is larger than what has been measured experimentally. For most of the features a Kendrick plot flags in a real sample, the CCS value, the authentic standard, and the toxicological profile do not exist. They are not in any database, because nobody has ever measured them.

What it means for analytical chemistry

The 70% problem is a stress test for the analytical techniques this blog is built around. We have the separation power. We have the mass accuracy. We have the CCS resolution. We have the suspect-screening and non-target-screening workflows. We are running out of databases.

Three questions are open right now:

What is the complete halogenated inventory in a real chlorinated drinking water — every regulated DBP, every suspect-screening hit, every Kendrick-plot feature that matches no database entry?
Of those features, which contribute meaningfully to toxicological load, and which are present at concentrations too low to matter?
For the features that matter, can we predict CCS values for compounds we have never measured, and use those predictions as an identification filter for the unknowns?

The first question is a non-target screening problem. The second is a toxicological prioritization problem. The third is a machine learning problem. That is the next post.