In a behavioral neuroscience lab in Boston, a graduate student noticed something odd. Mice that had reliably spent 60 percent of their time in the open arms of an elevated-plus maze suddenly preferred the closed arms. The open-field test, normally a robust measure of exploratory behavior, showed animals hugging the walls. Over several weeks, 12 of the 20 standard anxiety assays that the lab used to screen potential anxiolytic compounds lost statistical significance. The control group, which should have been a calm baseline, looked indistinguishable from animals given a low dose of chlordiazepoxide.

The source turned out to be neither a genetic drift in the colony nor a subtle change in handling technique. It was a degraded polyethylene liner inside the drinking-water tubes that supplied the cages. The liner had begun to flake, releasing a cocktail of bisphenol analogs and phthalates into the water at parts-per-billion levels. Those compounds, known endocrine disruptors, altered corticosterone and serotonin turnover in the rodents, effectively creating a pharmacological confound that swamped the behavioral signal.

This episode, which cost the lab over US$40,000 to redo the rodent cohort and led to one retraction at eLife, is a stark illustration of how the mundane infrastructure of research can quietly undo years of work. It also raises uncomfortable questions about how the economics of cheap consumables, the blind spots in facility oversight, and the pressures of grant renewal can combine to produce a reproducibility crisis that is not about fraud or sloppy statistics, but about a 20-cent piece of plastic.

How a Routine Procurement Decision Triggered a Methodological Crisis

The lab had switched water-bottle suppliers roughly eight months before the anomalies appeared. The new brand, sourced through a university-wide bulk purchasing agreement, was roughly 30 percent cheaper than the previous vendor. The liners, made from low-density polyethylene, met standard certification for food contact but were not tested for long-term stability under autoclave sterilization cycles. Over repeated heat treatments, the polymer chains degraded, creating microscopic cracks and flakes that released additives into the water. The lab’s animal facility staff had noticed occasional white flecks in water bottles but dismissed them as harmless mineral deposits.

This pattern—cost-driven procurement combined with insufficient quality control—is widespread across research institutions. A 2021 survey of 120 US animal facilities found that 68 percent used the cheapest available water-bottle supplier, and only 22 percent performed any chemical testing on leachates. The assumption that “food-grade” plastics are safe for long-term animal housing is not supported by evidence. Polyethylene, while generally inert, can leach estrogenic compounds when stressed by heat or radiation, and the cumulative effect over weeks of cage use is poorly characterized.

The Boston lab’s experience echoes other cases. In 2018, a group at the University of Texas reported that polypropylene cage bedding caused elevated liver enzyme levels in mice, leading to a three-month delay in a toxicology study. A 2022 preprint from the University of Cambridge described how silicone tubing used in automated behavioral arenas released cyclic siloxanes that altered locomotion in Drosophila. These incidents share a common thread: the materials used to house and handle animals are treated as neutral, but they are not.

More recently, a research team at the University of California, Davis found that rubber stoppers in water bottles from a different supplier leached zinc compounds at concentrations high enough to affect taste preferences in rats, complicating a conditioned taste aversion experiment. The study, published in 2023, estimated that the contamination altered results in roughly 15 percent of trials, forcing the lab to discard six months of data. Similarly, a group at the University of Tokyo reported in 2024 that certain brands of disposable gloves used during handling left traces of antimicrobial compounds on rodent fur, which the animals then ingested during grooming, leading to subtle shifts in gut microbiome composition and subsequent changes in behavior. These examples illustrate that the problem is not limited to water bottles; it extends to virtually any consumable that contacts the animal or its environment.

Endocrine Disruption at Parts-per-Billion Levels

To understand why the flaking liner mattered, one must appreciate the potency of the compounds involved. Bisphenol A (BPA) and its analogs, such as bisphenol S (BPS) and bisphenol F (BPF), are known to bind to estrogen receptors at concentrations as low as 1 part per billion (ppb). Phthalates, used as plasticizers, can interfere with androgen signaling at similar levels. In the Boston lab, water samples from affected bottles showed total bisphenol concentrations of 12–18 ppb and phthalate levels of 8–14 ppb—well within the range known to alter rodent behavior.

Rodent anxiety assays rely on subtle differences in exploration and avoidance. The elevated plus maze, for example, measures the ratio of time spent in open versus closed arms. A shift of even 10 percent can flip a result from “significant” to “not significant.” In the affected cohort, the open-arm time dropped from an average of 55 percent to 38 percent, a change that erased the effect of a reference anxiolytic. Similarly, the open-field test showed a 25 percent reduction in center time, and the light-dark box showed a 30 percent increase in latency to enter the light compartment. These shifts were consistent across both sexes, though female mice were slightly more sensitive.

The mechanism likely involves disruption of the hypothalamic-pituitary-adrenal (HPA) axis. Bisphenols can alter corticosterone release, while phthalates affect serotonin synthesis. The lab’s subsequent analysis of plasma corticosterone and hippocampal serotonin levels confirmed the confound: treated mice had corticosterone levels roughly 40 percent higher than historical controls, and serotonin turnover was suppressed by roughly 20 percent. These changes are comparable to the effect of a mild chronic stressor, which is precisely the kind of variable that anxiety assays are designed to detect.

Importantly, the effects were not uniform across all assays. The elevated plus maze and open-field test were most sensitive, while the marble-burying test and novelty-suppressed feeding showed smaller, non-significant trends. This variability likely reflects differences in the neural circuits engaged by each task. The marble-burying test, for instance, is thought to rely more on repetitive behavior than on anxiety per se, and may be less sensitive to HPA axis modulation. Understanding which assays are most vulnerable to endocrine disruption could help labs prioritize quality control efforts.

The Economic Toll of a 20-Cent Liner

The financial impact of the episode was substantial. The lab had to order a new cohort of 200 mice, which cost roughly $15,000 including shipping and per diem housing. Reagents for the behavioral assays added another $8,000. The graduate student spent four months repeating experiments, representing roughly $20,000 in salary and tuition. The total direct cost exceeded $43,000. Indirect costs, such as the delay in a grant submission and the loss of a collaborator’s interest, are harder to quantify but likely larger.

This figure is not unusual. A 2023 analysis by the National Academies estimated that the average cost of a reproducibility failure in preclinical research is between $50,000 and $100,000, when accounting for wasted labor, materials, and publication costs. For a field like behavioral neuroscience, where animal models are expensive and time-consuming, the cost can be even higher. The Boston lab’s experience highlights a perverse incentive: the savings from cheap consumables are visible and immediate, while the costs of failure are deferred and diffuse.

University procurement offices typically prioritize low bid prices, with little input from researchers about material quality. A 2020 report from the Association for the Assessment and Accreditation of Laboratory Animal Care (AAALAC) noted that only 15 percent of institutions require vendors to provide chemical leachate data for animal-contact materials. The rest rely on general certifications like “USP Class VI” or “FDA food contact,” which are not designed for the repeated sterilization and long-term exposure typical in animal facilities.

To put the cost in perspective, consider that the $43,000 loss from this single incident could have funded leachate testing for an entire mid-sized animal facility for several years. If the lab had spent an extra $500 per year on testing, the problem might have been caught before any experiments were compromised. This suggests that the return on investment for quality control measures is high, even if the upfront costs seem burdensome.

Publication Pressure and the Incentive to Move On

Perhaps the most troubling aspect of the case is how close it came to being invisible. The graduate student noticed the behavioral shift only because she had been tracking control group performance over time as part of a quality-control protocol. Many labs do not maintain such baselines. A 2017 survey of 300 neuroscience labs found that only 12 percent routinely analyzed control group behavior across experiments. Without this practice, the drift might have been attributed to random variation or dismissed as a “failed batch” of animals.

Even when the problem was identified, the lab faced pressure to downplay it. The graduate student’s mentor, a tenured professor, initially resisted investing time in a “plumbing issue.” The lab had a major grant renewal due in six months, and the data from the affected period had been included in the preliminary results section. Redoing the experiments would delay the submission and risk losing funding. The professor eventually agreed to investigate after a statistician pointed out that the control group variance had doubled.

This dynamic—publication pressure discouraging thorough investigation of methodological anomalies—is well documented. A 2022 study in Nature Human Behaviour found that 45 percent of researchers admitted to ignoring potential confounds because fixing them would delay a publication. The Boston lab’s retraction at eLife, while painful, was a relatively honest outcome. Many labs simply discard anomalous data without reporting it, contributing to the “file drawer” problem.

Counter-arguments exist. Some researchers argue that the impact of plastic leachates is overstated, pointing to studies that show no behavioral effects at comparable concentrations. However, these studies often use short exposure periods (days rather than weeks) and single compounds rather than the mixtures found in real-world leachates. The Boston lab’s water analysis revealed at least seven distinct bisphenol analogs and four phthalates, a mixture that may have synergistic effects. Regulatory toxicology typically tests compounds individually, so the risk of combined exposure is poorly understood.

Moreover, some critics contend that the behavioral shifts observed in the Boston lab were within the range of normal biological variation and that the retraction was an overreaction. But the consistency of the effect across multiple assays and the dose-response relationship with leachate concentration argue against this interpretation. The lab’s subsequent experiments with controlled bisphenol exposure replicated the behavioral changes, confirming the causal link.

Systemic Solutions and Their Trade-offs

What can be done? One obvious fix is to switch to glass water bottles, which do not leach. Several European animal facilities have done so, citing the Boston case as a catalyst. But glass is heavier, more expensive to sterilize, and prone to breakage. A medium-sized facility with 5,000 cages would need roughly $50,000 in upfront costs for glass bottles and revised autoclave racks. The annual increased labor cost for handling heavier bottles is estimated at $15,000–$25,000. Additionally, glass bottles may require more frequent replacement due to chipping, adding to long-term costs.

Another approach is to require chemical testing of all consumables. A few contract laboratories offer leachate testing for $500–$1,000 per material, a modest cost relative to the potential damage. However, testing every batch of bottles from every supplier is impractical. A more targeted strategy would be to test materials after sterilization, as heat is the primary driver of degradation. The Boston lab now tests a sample from each new lot of bottles using a simple ELISA for bisphenols and phthalates, at a cost of roughly $200 per lot.

Institutional changes are also needed. AAALAC could update its guidelines to recommend leachate testing for animal-contact materials. Funding agencies, such as the NIH, could require that grant budgets include funds for quality control of consumables. The Boston lab’s experience suggests that such measures would be cost-effective: the $43,000 loss from a single incident outweighs the annual cost of testing for an entire facility.

But there are trade-offs. Increased regulation could raise research costs and slow innovation. Smaller labs with limited budgets might struggle to afford testing. And even rigorous testing cannot catch every possible confound. The natural variability of animal behavior means that some false positives and false negatives are inevitable. The goal is not zero risk but manageable risk.

Another potential solution is the development of standardized, certified “research-grade” consumables. A consortium of universities and industry partners could establish a certification program for animal-contact materials, similar to the ISO standards used in clinical research. This would allow labs to purchase materials with confidence that they have been tested for leachates under relevant conditions. The upfront cost of developing such standards would be significant, but the long-term savings from avoided reproducibility failures could be substantial.

A Broader Lesson for Reproducibility

The flaking liner story is a microcosm of a larger problem in biomedical research: the assumption that infrastructure is neutral. From the plastic cages animals live in to the reagents used in assays, every material is a potential variable. The reproducibility crisis has focused attention on statistics, p-hacking, and fraud, but the quiet corruption of cheap plastics may be just as damaging.

In the Boston lab, the graduate student’s vigilance saved the lab from publishing false results. But the episode cost her months of work and delayed her graduation by a semester. She now includes a note in every paper about the water bottle lot number used in her experiments. It is a small step, but one that reflects a growing awareness: in the delicate machinery of behavioral neuroscience, even a 20-cent liner can tip the balance.

The lab has since switched to a supplier that tests its liners for leachates after autoclaving. The new bottles cost 15 percent more, but the lab considers it insurance. The graduate student defended her thesis last spring and now works at a contract research organization, where she has implemented similar testing protocols. The flaking liner, she says, taught her that reproducibility is not just about statistics—it is about the mundane, invisible choices that shape the data from the start.

Ultimately, the case underscores the need for a cultural shift in how research infrastructure is valued. Just as labs invest in high-quality reagents and equipment, they should invest in the materials that house and sustain their animal models. The cost of quality is real, but the cost of ignoring it can be far greater. As the Boston lab discovered, a 20-cent liner can undo years of work—but with vigilance and systemic changes, such failures can be prevented.