UPLC-ELSD Carbohydrate Question

[bmart]

I've been trying to develop a method for analyzing carbohydrates in beer utilizing an Agilent 1290 UPLC-ELSD set up with a waters Acquity UPLC Amide Column. The method has been following the application note found here: http://www.waters.com/webassets/cms/lib ... A64081.pdf on page 16. I made some modifications, basically extending the gradient time, because the line distances on the agilent system are longer than the waters equipment.

Thus far the separation has been good and I've gotten chromatograms similar to what water presents. My issue comes when I perform a matrix spike analysis. I've been spiking the sample with a mixture of glucose, sucrose, maltose, fructose, and maltotriose. When I do this spike with no dilution I get recoveries in the 150%-180% range for all of the sugars except sucrose which doesn't show up, weird I know. When I start diluting the sample 50:50 with the mobile phase, 30:70 Acetonitrile:water and 0.2% TEA, the sucrose shows up but I still get high recoveries. Then when I dilute the sample 4x with this mobile phase my recoveries go down to 40-60% range.

I've checked my math several times and had other people look it over too, we all agree that that is not the issue. Has anyone had good luck with developing a UPLC-ELSD carb method? Does anybody have any suggestions to alleviate this recovery problem? Thanks in advance.

[leadazide]

The ELSD had a fairly limited dynamic range.. Without claiming expertise with the ELSD could this maybe somehow be the cause??

[carlo.annaratone]

Did you take into consideration that ELSD response is non-linear? you must use a log/log calibration (were you plot the log of the response versus the log of the concentration)

[bmart]

Thanks carlo.annaratone. This was definitely the issue!

[JamieAlison]

Hello. I worked on developing and validating a similar method for the rapid analysis of monomers and oligomers in fermentation broth two years ago. The dynamic range for the ELSD is very short, especially when compared to the RI detector. What we also found was that the overall error in the method for the various analytes was consistantly in the 10-15% error range, depending upon the matrix and the analyte. For those of us used to the incredibly robust, accurate and precise nature of RI detection (with BioRad columns) this sounds like a shock. However, the overall resolution that we were able to achieve for some key analytes far outweighed the lower accuracy.

Hope this helps.
Jamie

[carlo.annaratone]

For bmart - what I do is to build a reverse calibration curve over the log/log response according to this formula:

log⁡(Concentration_analyte) = intercept + slope∙(log(Area_analyte )/log(Area_internalstandard.))

the justification for using an inverse calibration is found here:

Tellinghuisen, J. Inverse vs. classical calibration for small data sets. Fresenius. J. Anal. Chem. 2000, 368 (6), 585–588.

while the log/log response is described here:

L.E. Oppenheimer, T.H. Mourey, Examination of the concentration response of evaporative light-scattering mass detectors, J. Chromatogr. A. 323 (1985) 297–304.