ELISA Inter-Plate Variability: Same Sample, Different Results

The problem

To check reproducibility across plates, a researcher split each canine serum sample into two aliquots and ran them in parallel on two plates from the same lot of the Dog Beta-endorphin (BEP) ELISA, quantifying with a four-parameter logistic (4PL) curve. Two samples returned roughly double the concentration on one plate compared with the other, even though the serum was identical. A third repeated sample, by contrast, agreed closely between the two plates. The question was clear: with the same serum and the same lot, how could one plate read twice as high as another?

Snapshot of the data

The telling pattern is that the disagreement was sample-specific. Two samples diverged by about two-fold between plates, while a third repeated sample stayed consistent — a strong hint that the effect lived in individual samples, not in the plate or the curve.

Customer and sample identities are anonymized; values are illustrative of the reported pattern.

Why the same sample can read differently across plates

When the standard curve is valid but one sample varies between plates, the cause almost always sits at the sample level rather than in the kit. Three factors dominate: a matrix effect from undiluted serum, where sample-to-sample differences in composition shift the apparent signal; readings that land near the steep top or bottom of the 4PL curve, where a tiny change in optical density becomes a large change in back-calculated concentration; and small pipetting or handling inconsistencies that widen replicate scatter. A sample sitting comfortably in the middle of the curve, like the consistent repeat here, is buffered against all three — which is exactly why it agreed.

Our analysis

We started from what was solid. The standard curves on every run were clean and well-fitted, and the lot’s QC inter-assay CV was about 6.8% — comfortably inside the <10% target (up to 20% is acceptable). The capture-and-detection chemistry was therefore working as designed. The consistent third sample confirmed the assay reproduces well when a reading sits in the middle of the curve.

The two discordant samples told a different story. Their replicate CVs were elevated on both plates, and both had been measured undiluted — one high enough to sit near the upper plateau of the curve, the other low enough to give a weak optical density near the bottom. Both are regions where a 4PL curve is least precise. Crucially, a later re-test with a fresh kit reproduced the lower plate’s values, isolating the inflated readings to one particular run rather than to the kit or the lot.

Root cause

What we recommended

Because the assay itself was sound, the fix is to keep every sample reading in the reliable part of the curve and to remove handling variability:

• Run a full standard curve on every plate, so each run is validated independently and plates can be compared with confidence.
• Pre-test a few representative samples at several dilutions (for example 1:2 to 1:3) in the kit sample diluent, and choose the dilution that lands optical densities in the mid, linear portion of the curve while confirming acceptable linearity.
• Dilute serum to minimise matrix effects and pull high samples away from the upper plateau; avoid reporting values read at the very top or bottom of the curve.
• Standardise handling: centrifuge and mix samples, keep volumes and tip position consistent, avoid bubbles and contact with the well wall, wash thoroughly without overflow, never reuse plate seals, and keep pipettes calibrated.

Key takeaway

Frequently asked questions

What inter-assay CV should I expect from this kit?

Our target is an inter-assay CV below 10%, with up to 20% considered acceptable. The lot in this case returned about 6.8% in QC testing, so lot-level reproducibility was not the problem.

Same serum, two plates, very different values — is the kit faulty?

Not if the standard curves are valid and QC precision is in range. A sample-specific two-fold gap between plates usually points to handling, a matrix effect, or a reading near the extremes of the curve rather than a kit defect — especially when a co-run repeat sample stays consistent.

Why does a reading near the top of the curve matter?

A 4PL curve flattens at its top and bottom, so near those plateaus a small change in optical density translates into a large change in calculated concentration. Diluting the sample brings the reading into the steeper, mid-range portion where results are most reliable.

Should I dilute canine serum routinely for this assay?

Pre-test a few representative samples at several dilutions and pick the one that keeps optical densities mid-curve while confirming linearity. Validation targeted undiluted canine serum around 130 pg/mL, so samples reading much higher generally benefit from dilution.

Dog Beta-endorphin (BEP) ELISA Kit →

SKU: CNEB0155

Detection range	15.6–1000 pg/mL
Sensitivity	3.9 pg/mL
Sample types	Serum, plasma, tissue homogenates, cell culture supernatants and other biological fluids

Price: €749

View product →

Written by the Assay Genie scientific team — reviewed for accuracy by our PhD scientists. Customer and sample identities have been anonymized; data values are illustrative of the reported pattern and provided for educational purposes.