In elementary school, I was struck by a 1554 engraving showing residents of a European city emptying chamber pots onto the street. Gross, but also, to a kid, funny. The image oversimplifies pre-industrial sanitation. Cities relied on latrines, cesspits, and “night soil” collectors, who often sold the waste as fertilizer. But these systems buckled as cities grew.
Nineteenth-century London shows how badly. As the population exploded, so did the waste dumped into the Thames, which also supplied drinking water. The result was repeated cholera outbreaks starting in 1848, at a time when doctors blamed disease on “bad air,” not water. During an 1854 outbreak in Soho, physician John Snow mapped deaths and traced them to a single water pump. He then showed that neighborhoods served by a company drawing water from a dirtier stretch of the Thames had far higher death rates than those served by a company with a cleaner intake. Snow’s work did not immediately overturn miasma theory, but it helped show that disease could be traced to environmental exposure.
The real turning point came in 1858, when summer heat made the stench from the Thames so unbearable that Parliament itself demanded a fix. Engineer Joseph Bazalgette responded by building a sewer network that intercepted waste and carried it downstream, away from the city, reshaping London’s riverfront in the process.
That logic — get waste out and keep it out — defined sanitation for decades. But eventually, sewage stopped being just a problem to dispose of. It became a source of information. Public health officials began tracking poliovirus and, later, COVID-19 through wastewater. That approach, known as wastewater-based epidemiology, is worth a closer look before we get to food.
Sewage as a Source of Information
Wastewater-based epidemiology (WBE) uses sewage analysis to estimate population-level consumption of substances or biological agents, normalizing findings to a “per-capita mass load” based on daily flow and the population served.
In practice, WBE covers two territories: what people knowingly consume and what their bodies reveal without realizing it.
The first includes illicit drugs and medications. A 2005 proof-of-concept study estimated cocaine use in Italian cities through wastewater analysis, finding that surveys and crime statistics likely underestimated actual consumption. Researchers have since applied the same logic to alcohol, nicotine, and other drugs. For example, the Canadian Wastewater Survey has tracked several substances across major cities since 2018, revealing real regional differences. Not every substance cooperates, though. Some biomarkers degrade quickly, others are excreted unevenly, and non-consumption sources can muddy the signal.
Medications work similarly, but the focus is on gaps between sewage findings and official sales records. In the Netherlands, researchers compared sildenafil levels in wastewater from Amsterdam, Eindhoven, and Utrecht with dispensing data. They found that at least 60% of the levels could not be explained by legitimate prescriptions, suggesting online or unregulated use.
A more indirect use involves inferring disease prevalence. Some studies treat metformin as a proxy for type 2 diabetes rates, though not everyone with the condition is diagnosed or treated in the same way, and excretion varies between people. The same logic extends to chemical exposures, such as pesticides and parabens, as well as to antimicrobial resistance genes.
The best-known use, as mentioned, is infectious disease surveillance, which has long been used for polio and expanded dramatically during COVID-19. Sewage signals tend to arrive before clinical ones for a simple reason: infected people often shed viral genetic material in their stool before symptoms appear or a test gets ordered. That head start doesn't mean the RNA converts directly into a case count, though, because shedding varies by person and by the quirks of each sewer system.
From drugs to pathogens to chemical exposures, these applications share one idea: what a population excretes can reveal what's happening inside it. That raises a question WBE has only recently begun to explore: what if sewage could reveal not just what we consume as drugs or medicine, but what we eat?
The Traces of What We Eat
Published in the Proceedings of the National Academy of Sciences, a 2026 study introduces FoodSeq-FLOW (Food Landscape Observation in Wastewater). The tool extends an earlier method, which researchers developed to assess diet from individual stool samples, betting that the same approach could work on municipal sewage at a fraction of the cost of sequencing people one by one.
The team tested it on 183 sewage samples from 21 North Carolina treatment plants, representing about 2.1 million residents. Some samples were collected over time, while others were gathered simultaneously to compare communities. The tool detected dietary DNA for less than a penny per person, identifying 184 plant and 116 animal taxa associated with foods those communities commonly consumed.
To check whether the sewage signal reflected what people actually ate, the researchers compared it with individual stool samples from Durham residents. The two sources shared dozens of plant taxa, including grains, legumes, fruits, and spices, with abundances that closely tracked each other. The sequences aligned well with known foods: 98.9% of DNA matched edible animal and plant species. The rest likely came from environmental sources such as pollen or tree roots.
The tool also caught changes over time:
- Farmed fish such as Atlantic salmon and tilapia appeared steadily in urban areas year-round, matching their commercial availability, while other foods came and went with the seasons.
- Asparagus and blueberries appeared in spring and summer
- Brassicas and citrus appeared in late fall
- Turkey spiked every November and December.
Dietary signatures also tracked socioeconomic and demographic lines. Places with larger Asian and foreign-born populations showed more mung beans, chickpeas, mangoes, and coconut, foods associated with the dietary patterns of those communities and less typical of the local food landscape in North Carolina.
The balance between plant and animal DNA also varied by place, with cities like Durham and Charlotte leaning more plant-based. A handful of specific plants stood out as markers of place, with barley and hops, the backbone of beer, tracking with higher income and food spending.
Geography shaped seafood signals most of all. Fish made up just a third of animal DNA reads but accounted for 75% of taxa that distinguished one place from another. Inland cities leaned toward farmed salmon and imported tilapia, while coastal and rural communities relied more on local marine and freshwater fish.
The authors argue that sewage-based genomic profiling can aggregate individual dietary signals and scale to millions of people at low cost. That combination of low cost and the flexibility in what can be surveilled could support public health surveillance, targeted nutrition programs, and supply chain monitoring, complementing rather than replacing existing tools.
The caveats are real. DNA degrades at different rates depending on temperature, pH, and salinity; the data are semi-quantitative, so abundance doesn't map onto the amounts eaten; sampling misses households outside a plant's catchment, including those with a septic system; and environmental sources can add “noise” to the signal. Validation relied on a small set of stool samples from a single city, and the entire study was conducted in North Carolina, so researchers need to test the method elsewhere and replicate the findings independently.
The authors suggest testing whether marker combinations for soy, corn, and wheat could distinguish processed from unprocessed food. This refinement puts FoodSeq-FLOW side by side with the tools nutritional epidemiology has long relied on, but that are persistent sources of bias, like 24-hour recalls and food diaries.
If those results hold up in larger, more varied populations, FoodSeq-FLOW could offer something the field has long lacked. It could give researchers a way to check what people actually eat against methods that depend on memory and self-report, both notoriously unreliable. What we flush away, it turns out, may be one of the more honest records we have of what we actually eat — no recall bias required.
