ELISA Troubleshooting: High %CV at Low Standards

The problem

A researcher running a large clinical study (160+ samples) reported that the low end of the I-FABP/FABP2 standard curve was unusable. The upper points were clean, but the three lowest standards — 1.25, 0.625 and 0.3125 ng/mL — gave such high duplicate variability that concentrations could not be assigned. Because most of the clinical samples fell into exactly that low range, they could not be quantified and were slated for a costly repeat. The team had already pushed the top standard to 20 ng/mL (double the kit’s 10 ng/mL top point) to chase stronger readings, having noticed the low-end weakness early on.

Snapshot of the data

A representative plate: the top four standards are tight and monotonic, then %CV falls apart on the bottom three, where mean OD is barely above the blank.

On other plates the bottom points were worse still — once mean OD dropped to the blank, back-calculated values went out of range and individual wells even read negative after reference subtraction:

Values are illustrative and consolidated from the case to show the pattern; they are not a specific customer’s raw file.

Why it happens

At the bottom of any ELISA working range, net OD approaches the blank, so the signal-to-noise ratio is at its worst. A difference of a few milli-OD between duplicate wells — from mixing, an air bubble, an edge effect, or low-volume pipetting — is a small absolute error, but against a mean of 0.005–0.03 it is an enormous relative one. That is exactly what a %CV measures, which is why it explodes near the floor of the curve while the same technique looks flawless up top.

Two choices amplified the effect here. First, the plate was read in dual-wavelength mode (OD450 with a 620 nm reference subtracted). Near zero, the 450 and 620 readings are almost equal, so their difference can land below zero — producing the negative well values and “range?” flags seen above, and making %CV meaningless. Second, extending the top standard to 20 ng/mL (beyond the validated 10 ng/mL) changes how the 4-parameter logistic curve is anchored and weighted, which can degrade the fit at the very points that matter most at the low end.

Our analysis

The top four standards were tight (<10% CV) and cleanly monotonic, which tells us the reagents, coating and detection chemistry were all working. The failure was confined to the last three points, all sitting at or near the blank. Our QC team re-tested retained samples from the same lot and found no defect. The measurements existed only as OD450–620 (dual mode) with no single-wavelength raw data kept, so we could not separate true signal from reference-subtraction artifact — but the negative near-zero wells are the classic fingerprint of that artifact. Crucially, most of the clinical samples fell into this same low, noisy zone, which is why so many came back unquantifiable rather than just a handful.

Root cause

What we recommended

Key takeaway

FAQ

Why is my %CV huge only at the lowest standards?

Because their OD sits right at the blank. A few milli-OD of unavoidable well-to-well noise is a tiny absolute difference but a large percentage of a near-zero mean, so %CV inflates even when your technique is sound.

Can dual-wavelength (450/620) correction cause negative OD?

Yes. When net signal is close to zero, subtracting the 620 nm reference from OD450 can produce a value below zero. That is a mathematical artifact, not real negative absorbance — keep the raw OD450 as well.

Should I extend the top standard to get stronger low-end readings?

No. Going above the validated top point changes how the 4-PL curve is anchored and weighted and can worsen the fit at the low end. Stay within the kit’s specified range and adjust sample dilution instead.

My samples read below the lowest standard — what should I do?

Bring them up into the reliable part of the curve: run them neat or at lower dilution, or concentrate them. Anything still below the lowest reliable standard should be reported as < LLOQ, not as a quantitative value.

This article is based on an anonymized technical-support case and is provided for general educational purposes. Data shown are illustrative. Assay performance depends on sample type, handling and protocol adherence; always follow the current product datasheet. Specifications and pricing are subject to change — confirm details on the product page before ordering.