Grid cell replication failures get blamed on everything from animal strain to analysis code. A quieter culprit sits on the bench: the adjustable pipette. Two calibrations, one before and one mid-experiment, can quietly shift delivered volumes enough to move spike timing and change whether a finding replicates.

The Tension Between Replication and Craft

Grid cell results have a replication problem that is hard to pin on any single lab. The same hexagonal firing pattern shows up in some hands and blurs in others. The standard explanations, animal strain, electrode depth, analysis code, each absorb some blame. None fully accounts for the pattern of failures.

Pipette calibration rarely appears in methods sections. Most papers report the drug, the concentration, and the injection volume. They do not report when the pipette was last checked, or whether it was checked at all. That omission is small until you trace how a few percent of volume error propagates into a neural recording.

A related piece on this site argued that optogenetics papers depend on a virus batch that nobody standardised. The same logic applies to liquid handling. The variable is mundane, so it escapes scrutiny.

How Grid Cells Fire in Entorhinal Cortex

Grid cells sit in the entorhinal cortex and fire at regular intervals as an animal moves through open space. The firing fields form a hexagonal lattice that tiles the environment. The pattern was discovered in rats by Edvard and May-Britt Moser's lab, work that later earned a Nobel Prize. The cells integrate position, distance, and direction.

Grid cells have since been found in mice, bats, monkeys, and humans. The core property, a periodic spatial map, holds across species. What varies is the fine timing of spikes within each field, and that timing is where small perturbations show up.

A grid cell's firing rate depends on where the animal is, but the precise moment of each spike depends on the cell's membrane state. Anything that shifts that state, including a drug delivered at the wrong concentration, can move spikes by milliseconds. At the scale of a 125 ms theta cycle, milliseconds matter.

The Pipette as a Precision Instrument

Adjustable pipettes are mechanical devices with springs, seals, and a plunger. They drift with use. A pipette that delivers 100 microliters today may deliver 97 next month. Manufacturers typically recommend calibration checks every few months, but labs often stretch that interval.

Inaccuracy of a few percent compounds. A 3 percent volume error in a stock solution becomes a 3 percent concentration error in the working solution, which becomes a 3 percent error in the dose delivered to the brain. For a drug near the threshold of an effect, that is enough to flip a result.

The pipette entry on Wikipedia notes that measurement accuracy varies greatly depending on the instrument. That is a polite way of saying that a cheap pipette and a calibrated one are different tools. Labs rarely treat them as such.

Two Calibrations That Shape Results

The first calibration sets the baseline volume. It happens when the pipette is new or freshly serviced. The second calibration, if it happens, comes after a temperature change, a long experiment, or a suspicion that something is off. The drift between the two is the problem.

Temperature matters because liquid density changes with warmth. A pipette calibrated at 20 degrees Celsius delivers a slightly different volume at 25. Humidity affects evaporation from open reservoirs. Neither variable is typically logged in a grid cell paper.

Recorded data rarely includes calibration dates. A reader cannot tell whether the pipette was checked last week or last year. This site has noted that mapping how place cells fire taught neuroscientists to ignore the room. The same lesson applies to the bench: the instrument's history is part of the experiment.

From Preprint to Consensus or Rejection

When a grid cell finding appears as a preprint, independent labs attempt replication. If the effect holds, it moves toward consensus. If it fails, the failure gets attributed to unknown variables. Pipette records are usually unavailable, so the variable stays unknown.

A specific objection comes from labs that run automated liquid handlers. They argue that manual pipetting is inherently noisy and that the fix is to remove the human hand. That is fair, but most academic labs cannot afford a handler for every bench, and the handler itself needs calibration.

Consensus forms slowly without transparency. A finding that depends on a 3 percent volume difference will replicate in labs that happen to match that volume and fail in labs that do not. Neither lab did anything wrong. They just did not know what they were matching.

A related piece on this site argued that reproducibility checks stall where repositories stop accepting deposits. The same stall happens with wet-lab variables: there is nowhere standard to put them.

Practical Steps for Verifiable Replication

Log pipette calibration dates in the methods section. A single line, pipette X calibrated on date Y, gives readers a way to compare. It costs nothing and takes seconds.

Run a gravimetric check before each experiment. Weigh a dispensed volume of water on an analytical balance and compare it to the set point. This catches drift before it contaminates a dataset.

Report ambient temperature and humidity alongside the experiment. These are cheap numbers that contextualise the volume delivered. A lab at 25 degrees and a lab at 20 are not running the same protocol.

Share raw calibration data with preprints. Supplementary files can hold a calibration log without cluttering the main text. Readers who care will find it.

Adopt automated liquid handlers where the budget allows. They reduce hand-to-hand variance, though they introduce their own calibration schedule. The point is to make the variable visible, not to pretend it disappeared.

Case Study: A Grid Cell Replication Attempt

Consider a hypothetical but plausible scenario based on common laboratory practices. Lab A publishes a preprint showing that a certain NMDA receptor antagonist alters grid cell periodicity in mice. They report a 10 percent increase in grid spacing after injection. Lab B tries to replicate but finds no effect. Both labs use the same mouse strain, the same drug, and the same injection coordinates.

What differs is the pipette. Lab A calibrated their pipette two weeks before the experiment; Lab B had not calibrated in six months. A gravimetric check would have revealed that Lab B's pipette was delivering 5 percent less volume. That 5 percent difference meant the drug concentration in the brain was below the threshold needed to affect grid cells. The replication failed not because the science was wrong, but because the tools were not equivalent.

This scenario is not fiction; it reflects a class of errors that methods sections rarely capture. Without calibration logs, Lab B cannot know why they failed, and Lab A cannot prove their result is robust.

Trade-offs and Limitations

Focusing on pipette calibration is not a panacea. It adds time to already crowded protocols. A gravimetric check takes a few minutes per pipette, and logging calibration dates requires diligence. For high-throughput labs, the overhead may be prohibitive.

Moreover, calibration drift is just one of many sources of variability. Animal handling, electrode impedance, and analysis code all contribute. Singling out pipettes risks overlooking other factors. The goal is not to blame pipettes for all replication failures, but to acknowledge that they are a controllable variable that is often ignored.

Automated liquid handlers can reduce manual error, but they are expensive and require their own calibration. For many academic labs, the cost is prohibitive. A middle ground is to use calibrated pipettes and to log their status, accepting that some drift will occur.

Conclusion

Grid cell replication depends on many factors, but the humble pipette deserves more attention. By logging calibrations, running gravimetric checks, and reporting environmental conditions, labs can make their results more robust and more comparable. The alternative is to continue blaming unknown variables, when the answer may be sitting on the bench.