Estimation of uncertainty in size-exclusion chromatography with a double detection system (light-scattering and refractive index).

Size-exclusion chromatography (SEC) coupled with online laser light-scattering (LS) and refractive index (RI) detection provides an excellent approach to determine the molecular weights (Mw) of proteins by the "two-detector" approach. Mw is determined only at the maximum of a peak, using either peak heights or area ratio from the two detectors. However, proper calibration of the SEC/LS/RI system is critical to obtain high precision. Today, an essential part of any analysis is to evaluate the uncertainty associated with the method. Basically, it is possible to distinguish between factors related to signal nature, precision and those due to signal processing. Given the signal of interest is the peak height or area ratio from two detectors, the signal ratio uncertainty was calculated using the random propagation of error formula. In this case, the effect of signal correlation was evaluated to avoid the uncertainty overestimation. In the second case, the sources of uncertainty affecting analytical measurement were estimated with the information from the precision assessment. For this, two designs with two-factor fully nested were followed for each method. Finally, the contributions from various uncertainty sources related with calibration are also analysed in detail. There are in fact only three main sources of measurement uncertainty: intermediate precision, calibration and repeatability. Of these, method precision is always the greatest, regardless of approach. For all proteins and peptides studied, the Mw calculated using both methods are close to the theoretical results, independently of the design, but the contributions of individual terms to combined uncertainty depend on both the design and method used. For example, the combined uncertainty varied between 223 and 813.2 Da for carbonic anhydrase, although higher values were found for human insulin and ovalbumin dimer. Other considerations that can have a significant impact on the results are discussed. The reproducibility of the two methods versus that based on ASTRA software used as reference method was performed using the concordance correlation coefficient. The methods' reproducibility depends on the permitted losses in precision and accuracy.