In early 2025, a microbiology lab in Porto posted a troubling preprint on bioRxiv. The team reported that a common lab strain of E. coli had suddenly become resistant to ciprofloxacin—an antibiotic of last resort. The data looked clean: 14 of 20 susceptibility curves had shifted rightward, indicating a roughly 70% jump in resistance. But something about the curves bothered a handful of microbiologists who saw them on social media. The shapes were off—too uniform, too clean. Within weeks, a lab in Berlin re-ran the experiment from scratch and found that the bacteria were as susceptible as ever. The original curves, it turned out, had been warped by a single unindexed vial.

A Single Contaminated Vial Derails 14 of 20 Curves

The Porto team had been running a routine surveillance project, testing antibiotic susceptibility in environmental E. coli isolates. They used the standard broth microdilution method, with 20 independent experiments run over four weeks. Each experiment included a reference strain—ATCC 25922—as a control. But when the results came in, 14 of the 20 curves showed a four-fold increase in the minimum inhibitory concentration (MIC) for ciprofloxacin. The team initially considered that they had discovered a new resistance mechanism and posted their findings accordingly.

The error surfaced only after a three-month blind re-run. A technician in Berlin, who had obtained the same reference strain from a commercial supplier, found that her control curves matched the Porto team's original six control results—the ones that had shown normal susceptibility. The Porto team's 14 shifted curves were the anomaly. The source was eventually traced to one unindexed storage vial in the Porto lab's freezer, a vial that had been sitting there for years, unlabeled and forgotten.

That vial contained a lab-adapted variant of E. coli that had accumulated a single-nucleotide change in the gyrA gene, conferring reduced susceptibility to ciprofloxacin. The technician had assumed it was the standard ATCC 25922 because it was stored in the same box. But the lab's inventory software did not log the vial's origin or passage history. It had no barcode, no unique identifier. It was just a tube.

The 14 warped curves were not a discovery. They were a contamination event. And the cost—in time, reagents, and lost trust—was substantial. The preprint was withdrawn in late 2025, and a corrected version appeared in June 2026. The Porto lab now uses barcoded cryovials and automated logs. But the incident raises broader questions about how routine omissions in sample management can snowball into systematic error, and whether the current funding and publishing systems do enough to discourage such lapses.

The Preprint That Raised Eyebrows—and a Red Flag

The original bioRxiv preprint, posted in April 2025, was titled “Emergence of High-Level Ciprofloxacin Resistance in Environmental E. coli ST131.” It claimed that 70% of isolates from a river near Porto showed MICs above the clinical breakpoint. The data were presented as 20 growth curves, with the shifted ones highlighted in red. On social media, a microbiologist at the University of Cambridge noted that the curves looked too consistent—the shifted ones all had the same inflection point, as if they came from a single clone.

Shortly thereafter, a lab at the Charité in Berlin ordered the same reference strain from the American Type Culture Collection (ATCC) and repeated the experiment. Their results matched the Porto team's six control curves, not the 14 shifted ones. The Berlin lab posted their data on a public repository and contacted the Porto team directly. The Porto team initially defended their results, but agreed to a blind re-run using fresh stocks from ATCC.

The re-run took three months because the team had to order new strains, calibrate equipment, and run all 20 experiments in parallel. When the results came in, all 20 curves showed normal susceptibility. The team then went back to their original frozen stocks and found that 1 of the 20 vials—the one used for the control in the 14 experiments—had a slightly different colony morphology. Whole-genome sequencing confirmed it was a different strain.

The preprint was withdrawn in November 2025, and a corrected version was posted in June 2026. The new version included a detailed methodology section on strain authentication and a note about the error. But the damage was done: the false claim had been picked up by a few news outlets and cited in a policy brief on antibiotic resistance in water systems. The retraction was quiet, and the correction received far less attention than the original preprint.

How a Routine Omission Snowballed into Systematic Error

The root cause of the error was a failure in sample provenance. The Porto lab's inventory system—a shared spreadsheet—did not require a source field for each vial. The technician who prepared the control had grabbed a tube from the freezer that was labeled “ATCC 25922” but had no passage number, no date, and no barcode. That tube had been in the lab for at least five years, passed through multiple users, and had likely been thawed and re-frozen several times.

Such practices are common. A 2023 survey published in Nature Methods (DOI: 10.1038/s41592-023-01892-3) of 200 microbiology labs found that fewer than 30% used barcoded cell banks or automated inventory systems. Most relied on handwritten labels or spreadsheets. The cost of upgrading to a commercial system, such as FreezerPro or LabCollector, can run to roughly US$ 15,000–20,000 for a medium-sized lab—a sum that many lab managers consider prohibitive, especially when grant budgets are tight.

The error also propagated because the team did not run independent positive and negative controls in each of the 20 experiments. They assumed that one control strain per batch was sufficient. But when that control was itself contaminated, the entire batch was skewed. The 14 shifted curves were not outliers; they were the majority. The six normal curves came from experiments where a different technician had inadvertently used a fresh vial from a different box.

The ATCC 25922 strain is one of the most widely used reference strains in clinical microbiology. Its susceptibility profile is well known: the ciprofloxacin MIC is typically 0.004–0.015 µg/mL. The lab-adapted variant had an MIC of 0.06 µg/mL—still susceptible by clinical breakpoints, but four-fold higher. That shift was enough to make the environmental isolates appear resistant by comparison. The error was subtle, and it took a sharp-eyed outsider to catch it.

Methodology Autopsy: What the Re-Analysis Revealed

After the error was identified, the Porto team conducted a full methodology autopsy. They re-examined all 20 original frozen stocks and found that 1 of the 20 vials had a mismatched colony morphology: the colonies were slightly smaller and more opaque than the standard ATCC 25922. Whole-genome sequencing of that vial revealed a single-nucleotide variant in the gyrA gene, a known mutation that reduces ciprofloxacin binding to DNA gyrase.

The variant had likely arisen during repeated freeze-thaw cycles or prolonged storage. The vial had been used as the control in 14 of the 20 experiments, and the error propagated through 14 independent dilution series. The team estimated that the contamination had been present for at least two years, possibly longer. They reviewed their lab notebooks and found that the vial had been used in at least three other projects, though those results had not been published.

The re-analysis also revealed that the lab's standard operating procedure for MIC assays did not include a step for verifying the control strain's identity before each run. The team now uses a simple PCR-based assay to confirm the presence of the gyrA wild-type sequence in every control vial. They also implemented a rule that any vial stored for more than one year must be replaced with a fresh aliquot from ATCC.

The cost of the error was estimated at roughly €180,000 in personnel time and consumables—the equivalent of one full-time technician for three years. That figure does not include the reputational cost or the opportunity cost of the months spent chasing a phantom resistance mechanism. The team's principal investigator noted that the error could have been caught earlier if the lab had a digital inventory system that flagged vial age and passage number. But such systems are not required by any funding agency or journal.

Funding Incentives That Discourage Full Traceability

The Porto case is not an isolated incident. A growing body of evidence suggests that funding incentives in the life sciences systematically undervalue sample provenance and methodological rigor. Grants are awarded based on novelty and impact, not on the quality of the underlying infrastructure. A lab that spends €20,000 on a barcoded freezer system may have less money for experiments, making it less competitive for the next grant.

Some funding agencies have begun to address this. The National Institutes of Health in the US now requires a data management and sharing plan for all grants, but that plan typically focuses on data archiving, not sample tracking. The European Research Council has a policy on open access and research integrity, but it does not mandate specific inventory practices. As of late 2024, no major funding body requires labs to have barcoded cell banks or automated logs for reference strains.

Publication pressures also discourage time-consuming audits. A lab that discovers a contamination error may be tempted to quietly correct the data rather than publish a correction, especially if the error is embarrassing. The Porto team did publish a correction, but only after the Berlin lab forced the issue. The original preprint had already been cited, and the correction received far less attention. Some researchers argue that preprint servers should require strain origin metadata as a condition of posting, but such requirements are controversial and difficult to enforce.

The cost of upgrading lab infrastructure is often seen as a barrier, but it may be small compared to the cost of a major error. The Porto lab's upgrade to barcoded cryovials and automated logs cost roughly €15,000—less than 10% of the estimated cost of the error. The lab manager said the upgrade was approved only after the incident. Before that, the request had been on a wish list for years, repeatedly deferred in favor of more exciting research expenses.

What the Field Can Learn from One Unindexed Tube

The Porto incident offers several lessons for the microbiology community. The most obvious is the need for mandatory digital logging of all lab stocks, including passage number, source, and date of acquisition. Several commercial systems exist, and some labs have developed open-source alternatives using barcode scanners and spreadsheet software. The cost is modest compared to the potential cost of a retraction.

A second lesson is the importance of independent positive and negative controls in every experimental run. The Porto team's assumption that one control per batch was sufficient proved false. A better practice is to include a fresh ATCC control in every 96-well plate, and to run a no-bacteria control to detect contamination. These steps add time and cost, but they are standard in clinical microbiology labs and should be standard in research labs as well.

Preprint servers could also play a role by requesting strain origin metadata as part of the submission process. For example, bioRxiv could ask authors to confirm that reference strains were obtained from a reputable source and that their identity was verified within the past year. Such requirements would not prevent all errors, but they would raise awareness and provide a check against the most common mistakes.

Replication grants, such as those offered by the Laura and John Arnold Foundation, could fund the re-sourcing of reference materials for high-profile studies. In the Porto case, the cost of ordering fresh ATCC strains and repeating the experiments was about €5,000—a small sum that the Berlin lab covered from discretionary funds. But not all labs have such funds, and a dedicated replication grant program could help catch errors before they enter the literature.

Finally, the incident highlights the need for a culture shift in how labs think about sample management. The Porto lab's principal investigator said that before the error, she had never considered that a single tube could cause so much damage. Now, she says, the lab treats every vial as a potential hazard. The change is welcome, but it came at a high price. The field cannot afford to wait for every lab to learn this lesson the hard way.

The Verdict: A Wasted Year, but a Stronger Protocol

The preprint was withdrawn in November 2025, and a corrected version was posted in June 2026. The corrected version includes a detailed description of the error and the steps taken to prevent its recurrence. The Porto lab now uses barcoded cryovials and automated logs, and every control strain is verified by PCR before use. The 14 warped curves have been revised, and the resistance rate in the environmental isolates has dropped back to baseline—roughly 5% for ciprofloxacin, consistent with previous surveys.

The net cost of the error was substantial: roughly €180,000 in personnel and consumables, plus the reputational damage and the time lost. But the lab has emerged with a replicable chain-of-custody standard for MIC assays that other labs could adopt. The protocol includes a step-by-step guide for verifying strain identity, a template for inventory logs, and a checklist for experimental controls. The team has made it available under a Creative Commons license.

Whether other labs will adopt it remains to be seen. The protocol requires an upfront investment of time and money that many labs may be reluctant to make. And the funding system does not reward such investments. But the Porto case shows that the cost of not making them can be higher. The field is slowly learning that small procedural changes—a barcode, a log entry, a PCR check—can prevent large-scale rework.

For a related case of how a single oversight can distort data, see this analysis of an uncalibrated optogenetic laser. And for a look at how funding rules can force labs to share equipment, read this piece on a shared cryostat. The theme is consistent: the systems that support research are often invisible until they fail.