How do resistance indicators in sewage microbiota mirror clinical antibiotic use?

Host taxonomy splits the sewage resistance signal

The study asks whether antibiotic resistance gene (ARG) abundance in sewage influent can be read as an epidemiological indicator of community antibiotic use. Its central result is that the answer depends on which bacteria carry the gene. ARGs linked to Enterobacteriaceae, the family that includes common fecal enteric bacteria such as E. coli, tracked antibiotic usage most closely when no lag was applied. ARGs predominantly encoded in Pseudomonadaceae, an ecologically distinct family that can persist in sewer environments, instead showed positive correlations with usage at increasing lag, most prominently at two months.

That split matters because a summed drug-class ARG abundance, the usual way sewage resistance is aggregated, mixes genes from hosts with different lifecycles in the sewer. The authors argue that this mixing is one reason earlier work found only weak relationships between antibiotic use and resistance in sewage. By correlating individual gene clusters rather than summed class abundances, they could separate a near-immediate fecal-linked signal from a delayed environmental one.

One plant, two samples a week, for a year

The evidence comes from composite influent samples collected twice weekly for roughly one year, August 2020 to September 2021, at a single small conventional wastewater treatment plant of about 3.5 million gallons per day. Samples were deep-sequenced, averaging 29 gigabasepairs or 197 million reads per sample, for a total of 3,248 Gbp of Illumina data. Estimated coverage averaged 0.74, with a range of 0.59 to 0.90.

Antibiotic use was measured separately, as outpatient prescriptions from 48 facilities covering 2017 to 2023, spanning 13 major antibiotic classes and 27 subclasses. Prescriptions were expressed as mean daily prescriptions per 10,000 visits, assigned to the sewershed by patient home address. The five-year prescription record is what makes lag testing possible: the year of sewage sampling sits inside a much longer usage series, so correlations can be shifted backward in time by one, two, or three months.

The sampling year overlapped COVID-19 interventions, including school closures and a shift to online instruction. The authors note that overall outpatient prescription rates during the sampling months were depressed relative to earlier years, and that crAssphage, a human fecal marker, fell from November to January as university classes moved online. These are features of the study period, not corrections applied to the main correlation analysis.

Why a lag is plausible in the sewer

A lag between a prescription and a resistance signal requires a mechanism, and the paper offers several that operate inside the sewer rather than in the patient. Sewage influent is a mixture of human excreta, greywater, groundwater inflow and infiltration, and sewer-associated biofilms and sediment. Sediments and biofilms can hold resistant organisms, resistance genes, and antibiotics, and sloughed material from them contributes to the signal reaching the treatment plant.

For trimethoprim specifically, the paper points to persistence: nearly all ingested trimethoprim is excreted unchanged, its environmental half-life is reported as several days to several months, and adsorption accounts for more than 99% of trimethoprim removal in activated sludge. The authors also raise an alternative: susceptible populations in the sewer could die out and be overtaken by resistant Pseudomonas. Both explanations are presented as possible drivers of a selection delay, not as a demonstrated causal chain.

Drug classes differ in whether they lag

The lag pattern is not uniform across antibiotic classes, and the counts of correlated genes make the contrast concrete. Penicillin prescriptions correlated directly with more than 49 ARGs but with only 6 when usage was lagged. Macrolides ran the other way: only two corresponding ARGs were positively correlated without lag, while 46 were positively correlated at a two-month lag. Antibacterial folate antagonist combinations correlated with two ARGs without lag and 44 at a one-month lag. The authors report this contrast as statistically significant using a two-tailed Fisher's exact test with p < 0.0001.

The authors connect the unlagged beta-lactam result to the environmental instability of beta-lactam rings, which are easily hydrolysable, so a direct association is what they expected. One exception they flag is second-generation cephalosporins, almost entirely cefuroxime prescriptions, which were most prominently associated with resistance at a two-month lag. The authors state plainly that they cannot know whether this reflects real differences in cephalosporin selection pressure, and they note cefuroxime was the most scarcely prescribed cephalosporin in the data.

Trimethoprim links to Pseudomonas after adjustment

The strongest host-specific test uses nested statistical models on Pseudomonas metagenome-assembled genomes, or MAGs, which are genomes reconstructed from the metagenomic reads. The base model fits a seasonal cosine term; successive models add lagged sulfamethoxazole/trimethoprim prescriptions, then influent flow, crAssphage abundance, and a weekday indicator for Monday versus Friday sampling. Influent flow is included because it relates to pipe shear stress on biofilms, which the authors expect to affect release of biofilm-associated microbiota.

After accounting for seasonality and these sewage characteristics, the models link Pseudomonas MAG dynamics with lagged sulfamethoxazole/trimethoprim use. Chemical screening supports the plausibility of that pairing: among the antibiotics detected in the sampled wastewater, trimethoprim was the most prevalent, with a peak area 6.4-fold greater than the next most prevalent, ciprofloxacin. Azithromycin, sulfamethoxazole/trimethoprim, and ciprofloxacin were highlighted as both commonly prescribed and detected in sewage. Beta-lactams other than imipenem were not included in the suspect screening because of their environmental instability.

What the clinical lens misses

The trimethoprim–Pseudomonadaceae result carries an implication the authors draw out explicitly: clinical monitoring and environmental impact can diverge. Trimethoprim is not generally considered a clinically relevant therapy for P. aeruginosa, so resistance in Pseudomonadaceae is not surveilled in clinical settings. Yet in this dataset it is the host group whose resistance signal appears after a lag and survives adjustment for sewage conditions and seasonality.

The paper also notes that it did not observe major Pseudomonas blooms, unlike the extreme blooms reported elsewhere, and suggests infrastructure differences between sites could explain that. This is a caution about generalizing the host-lag structure: the authors describe their results as derived from in-depth analysis of a single treatment plant and community, and expect environmental connectedness and system design to shape resistance indicators elsewhere.

Where the host-lag reading holds

The practical priority this paper establishes is a choice of indicator, not a single monitoring number. If the goal is to track community antibiotic use through sewage, ARGs whose predicted hosts are Enterobacteriaceae carry the near-term signal, while Pseudomonadaceae-associated ARGs carry a delayed one and should be read against usage one to three months earlier. Aggregating ARGs by drug class alone would blend the two and obscure the timing.

The boundary is equally specific. The correlations are associations between prescription records and gene abundances, not demonstrations that antibiotic use caused the resistance levels observed, and the authors frame the selection-delay mechanisms as potential drivers. The lag analysis considered up to three months, a horizon borrowed from prior seasonal work. And the whole structure rests on one plant and one community during a pandemic-disrupted year, so the host-lag split is a hypothesis to be tested in other sewersheds rather than a calibration that transfers directly. What the paper does establish is that the host ecology of sewage microbiota is the variable that decides whether a resistance indicator reflects recent prescribing or a delayed environmental response.

Sources used in this answer