I'm doing some ELISA development, and I'd like some justification/best practices for background correction. I'm using a horseradish-peroxidase (HRP) detection system along with a TMB substrate and an acid-based stop solution, which is quite common. When TMB is cleaved by HRP, it produces a blue (maximum absorbance (Amax) of 650 nm) product. The addition of the acid not only stops the reaction for endpoint detection, but it amplifies the signal two- to three-fold, and turns the solution yellow, so that it should be read at its Amax of 450 nm. The acid itself supposedly has an Amax of 540 nm, which is why you should use that wavelength for background correction, subtracting it from the A450. Other sources claim the A540 reading is supposed to correct for "optical imperfections" in the plate.
Using either of these explanations, one would assume that the A540 would be relatively uniform across the plate, independent of the A450 of the same well. However, in my experience, that is not the case at all. My blanks and negative controls are averaging about 0.041 absorbance units (AU) at 540 nm, with an average A450 of about 0.065. However, an A450 of 0.9 gives an A540 of 0.051, 1.4 gives 0.6, 2.0 gives 0.73, and 3.5 gives about 0.14. The absorbance spectrum (Fig. 2) of the substrate appears to be quite minimal at 540 nm, so why am I seeing this dose-dependent increase as the A450 increases?
More practically, how should I do background correction? Currently, I'm subtracting each individual well's A540 from its A450. Should I continue this practice, or should I determine an average plate background at 540 nm using my blanks and negative controls, and subtract that value from each individual A450?