Optimizing MS parameters with an unstable signal

[quantonium]

My plan is to perform do a Box-Behnken response surface design optimization of ESI ion source parameters (sheath gas, voltage, etc.). However, despite much testing, including people from Thermo, we still have some fairly large intensity drops that can happen randomly (one run to the next, one day to the next..). As signal intensity (area of my analyte) was going to be my response, this would be problematic. An internal standard works for quantitation, but not absolute intensity comparisons..

Can anyone think of a smart way to circumvent this problem?

[Multidimensional]

The problem with the instrument must be fixed before any useful optimization steps can be taken. It is unscientific to collect data on a system that has known problems. Find the causes, repair, test, then monitor before running any samples.
Note: LC-MS system which have poor or intermittent reproducibility issues need attention from a very experienced chromatographer. Start troubleshooting at the HPLC system (with no column installed. Install a calibrated restriction capillary in place of the column to provide the required > 40 Bars system back-presure). Verify the in-line degasser is clean, functioning perfectly (if it is more than 3 years old, it may be due for professional service), bottles are clean, solvent pickup frits are new, solvents of the correct grade and purity before starting. Verify the pump is capable of flowing solvent, under normal pressures at the stated flow rates (verify with a graduated cylinder and timer) and has been fully cleaned and serviced. Same with Autoinjector/Sampler (problems with the sampler may result in changes in sample injection volumes) and other modules. Run instrument performance verification methods to insure the work was done correctly. Once the entire HPLC system has been verified, move on to the MS source, valves, tubing, cell etc. The software settings used for the method, data acquisition also need to be checked. Use one or two pure standards to check initial values, then full calibration mixes after things look good.

[quantonium]

Sadly, all of the above has already been extensively looked at, and we're still experiencing issues. We're still attempting to fix it though, so let's hope it works out. Anyway, thanks for the answer!

[bcd_GCLCMSMS]

Are you optimizing the source for absolute signal (sensitivity) or stability? Many people optimize the source for maximum absolute signal, but in my experience that doesn't always provide the most stable signal over multiple injections.

Many large clinical laboratories optimize based on the total analytical error principle, which includes accuracy (recovery) and standard deviation over 7 or more measurements. The standard deviation component is weighted heavier than recovery in the equation. Mass spec. instruments traditionally have absolute signal RSDs of 2-8% on repeated injections of the same sample. If the RSD is greater than this, then there is likely a hardware problem or large adjustment problem.

[quantonium]

That's interesting. In any case, the idea was to optimize for absolute intensity, as I'm working with extremely limited analyte concentration. Regardless, even optimizing for RSD wouldn't really work with our MS at the moment, as the signal instability seems to be completely random. I'm might get RSDs of 50 % one day, and 5 % the other.

[James_Ball]

That's interesting. In any case, the idea was to optimize for absolute intensity, as I'm working with extremely limited analyte concentration. Regardless, even optimizing for RSD wouldn't really work with our MS at the moment, as the signal instability seems to be completely random. I'm might get RSDs of 50 % one day, and 5 % the other.

Are these variations in a clean standard matrix or with a mixture of standards and samples?

Could it be that something in the sample matrix causes the change in sensitivity? If so you may need to alter or clean the matrix to remove the instability. Dilution is one way to potentially remove matrix, but another is to use the switching valve on the instrument to reject everything eluting from the column except in a narrow window around the targets of interest. I had to do this when analyzing samples of extracts from tobacco.

[quantonium]

It's a good idea, but sadly, that also has been tested extensively. The samples are relatively simple (water+MeOH+oxysterols+a few other things), made even simpler by an online trap and elute setup using two pumps. Even so, the issue has been seen with three people already, with widely different setups and samples.