ELSD - Lesson learned!

[juddc]

Good afternoon!

Thought I'd share what I think is a useful lesson that I learned this week while quantifying caffeine using a photodiode array and ELSD in series.

The samples: 3 Prototype creams with significant concentrations of caffeine to be quantified.

The goal: Quantify the caffeine as requested and provide a fingerprint of each sample identifying as many of the other major components as possible (excipients, preservatives, other actives, etc...).

The result: Calibrations for both detectors (6 points from 0.25-2X dose) RSQ > 0.999. ELSD calibration was quadratic, of course. System suitability (area) for PDA = 0.34%; for ELSD = 1.36% (N=6).

PDA: Sample recoveries were perfectly on target: 99.5-101% recovery on all 3.
ELSD: Sample recoveries ranged from 80.3-81.1%

What HAPPENED??

After a few moments of head scratching, it came to me:

1. The caffeine standards were dissolved in MeOH-H2O (1:1) whereas the creams were dissolved in neat MeOH in order to pull as much of the rest of the sample (preservatives, other actives, lipids, etc...) into solution and to precipitate any water soluble thickeners (carbomer, etc) that may be present.

2. The caffeine peak is the earliest eluter in the method.

3. My system volume is relatively small compared to my injection volume.

4. The ELSD is a non-linear detector.

What happened, I believe, was that there was slight band broadening with my samples that was not present with the standards. This was tough to see at a quick glance because my sample runs were 30 minute gradients whereas my standard runs were 5 minute isocratic runs. The sample gradient has a 5.5 minute isocratic delay at the start in which the caffeine elutes, so I can get away with that. Close inspection of the sample chromatograms revealed a slightly rounded peak for caffeine! That band broadening didn't effect the PDA results to any great degree, but because the ELSD is non linear, it really did a job on those results. Peak purity, by the way, was fine and consistent between the standards and the samples.

Preparing some standards in neat methanol and recalibrating with slightly broader peaks on the ELSD confirmed the theory - recoveries mirrored the PDA results perfectly.

Dropping the injection volume from 10ul to 5ul for samples and standards also brought peak shapes more in line.

I'm considering incorporating SPE into this particular sample prep should these prototypes move much further forward.

Lesson learned! Hope you don't mind my sharing my goof du-jour!

[Noser222]

That was interesting...so many of us post our problems here...it's nice to see someone post a solution once in a while

[AA]

You also may not know this, but caffeine is considered to be a semi-volitile in terms of ELSD detection. If you hace the drift tube temperature too high, your caffeine peak height will be reduced or dissappear compleatly (one of the hard ELSD I learned). As well, you are quite correct in saying ELSD in non-linear. It is not just your detector, the technique itself is non-linear., the calibration curves fit very nicely to quadratic equations though.

[juddc]

Thanks for the comments!

Loss due to analyte volatility in the detector was not a real issue here and would have been consistent from standards to samples regardless. I'm fairly convinced that differential band broadening between samples and standrds was the culprit, expecially given the results from the second experiment. The signal to noise ratio for caffeine was actually quite good despite a mobile phase composition of ca 90% Aq 0.02M NH4Ac at that point. At the 1X dosage, I'd estimate a s/n at >500:1.

I wasn't primarily interested in quantifying the caffeine with the ELSD as I had the PDA to do that, but I thought it would be interesting to see how well quantitative data correlated between the two detectors using "real world" samples. I was using the ELSD primarily to help generate a more complete chromatographic fingerprint of the sample in order to expand the potential number of analytes to be considered quantifiable as an aid to further formulation work.

I just wanted to point this issue out to other users of ELSD's so that they avoid the issue I ran into...or if they DO run into it, they'll be able to identify it without making themselves crazy!

Sample preparation with creams and lotions can sometimes be tricky. Sometimes you can bend the rules (as long as you're aware of which rules you're bending!), sometimes you can't!

[HW Mueller]

If I follow this correctly then the real lesson to be learned here is to calibrate your system under the same conditions which are to be used in sample runs? Calibration should be "teaching" the equipment of how the analyte will respond in your sample? If there have to be some diffs between standard and sample you can add an internal standard (IS) to give you an indication of what the effects of the diffs may be.

(To obviate replies about the negative aspects of IS: One can/should use the external standard method in parallel).

[juddc]

Agreed...

Further, I'd think that anything that might cause peak width / height variations for any given analyte (gradient control issues, pump flow variations, etc...) would have a greater impact on an analyte quantified by ELSD.