Non-linear calibration GCMS

[jdlh199]

I have an assay by GC-ECD from 2 to 100ppm that we are moving to GCMS with a lower calibration range from 0.1 to 10ppm. By GC-ECD the calibration curve is perfectly linear. But by GCMS I am seeing the below.

So my question is, is non-linear calibration "normal" in GCMS work. How can I convince our client that is acceptable to use this calibration curve. Or is there a problem with our method/instrument that needs fixing!

[Peter Apps]

This is not so much to do with GC-MS vs GC-ECD as with the change in range of calibration, which is taking you down into an area where adsorbtive or absorptive losses are significant i.e. below 2 ppm.

Are you seeing peak tailing at the bottom end of the calibration ?

You need to look carefully at the deactivation of the inlet, and any glass wool etc that is in there, and the state of the column. You might be getting losses during sample prep or making up the standards also.

I would recommend having your calibration points spaced evenly along the line.

Peter

[Don_Hilton]

That is a lovely calibration curve.

The way to convince your client that this is good would be to give some data showing the determination of LOD, bias, and precision of your technique - Take a standard with a concentration such that the signal to noise is about 20:1 or 30:1 and make five to 7 injections. Use a t-test to get the confidence interval. Below this confidence interval, you can not be sure that you are not integrating noise peaks. But above this confidence interval the peaks are real. RSD's along the curve - and I would do this with some spiked samples - will indicate the quality of data at those levels.

What I have given is off the top of my head. If you look in the US EPA methods, they specify a formal - and slightly more involved method of looking for the LOD and LOQ.

Drawing the curve is a good start. Showing that numbers are reliable along that curve is the tough part.

Let me know if this helps.

Don

[jdlh199]

Thanks for the tips.

I'm not sure it's analyst error since its the same analyst who has been preparing the ECD samples for a year or so now. Would inlet deactivation/adsorption affect the lower concentrations significantly more than the upper concentrations? I guess it could, but I could probably rule that out by running a high conc. curve at the same concentrations as the ECD curve. But, ignoring the lower end of the MS curve above, the standards at 2, 5, and 10 by themselves are not linear by GCMS, but they are by GC-ECD.

Is there anything at the MS in terms of instrument problems that could cause this? I'm sure we don't have a leak since that's checked before running, as is MS tuning. What about an ionization effect that occurs in MS but not in ECD?

I guess we'll run another compound to show that one is linear, maybe run another curve with samples/spiked samples/QC's etc. and look at accuracy/precision etc. I'm just used to working in linear...!

The standard level distribution is due to the fact that we usually use a 1/x weighted linear curve, which always seems to work better when standards are at 1, 2, 5, 10, rather than 2, 4, 6, 8, 10, for example.

[Peter Apps]

Ad/absorption has a greater impact with small quantities since it tends to affect only a limited and fairly consistent mass of analyte. Rather than losing 10% of the peak at all levels, you lose say 100 pg, which is 10% of a 1 ng peak, but only 1% of a 10 ng peak and an irrelevant and undetectable 0.1% of a 100 ng peak.

Another possibility - are you making your calibration standards by serial or by point dilution ? If by serial dilution small volumetric (or gravimetric) errors at each step can accumulate as you work down from the highest to the lowest concentrations ?

Who mentioned analyst error ?

Peter

[gpronger]

Generally speaking the mass-spec should be as linear as the ECD (this is of course dependent upon the particular model GC and particular model GCMS). Your calibration looks like a typical curve, that is non-linear due to active sites somewhere in the system.

In my experience, I would look at issues with the inlet or active sites at the front of the column.

Is the inlet liner the same?
Are the inlet conditions the same?
Is the column in the GCMS in as good of shape as the GC?
Is there obvious differences in chromatography?

Greg

[Don_Hilton]

The non-linearity could be activity -- or it could be that the noise in the base line has a roll such that it looks like it is part of the peak - and the upper half of the roll tends to be included in peak area. Depending on the instrument and compound being analyzed, 0.1 ppb gets you into the neighborhood of this problem. The LOD calculation will sort that our really fast. I have noticed that the eye is able to very reproducably pick out a small peak in the baseline - even when there is nothing present. And, the RSD tends to be pretty good. (But, not an RSD you want to report.)

[jdlh199]

Well it's been a while, but I figured I'd come back to this problem. So here is my update... If anybody has additional tips based on this, I'm really interested in hearing them.

So I prepared an extended calibration curve and aliquoted into two sets of vials, one for ECD analysis, and one for MS analysis. Everything up until the point the column enters the detector is the same, i.e. all GC program settings, all GC hardware, column, liner etc etc. And the columns are of equivalent age/condition.

Here is my GCMS curve: http://i41.tinypic.com/2d0ifcm.jpg

Here is my GC-ECD curve: http://i42.tinypic.com/xpzhp5.jpg

So MS is consistently non-linear, no matter if I'm looking at higher or lower concentration ranges. This test above was WAY over detection limits, so integration is not an issue. The ECD method is always linear and always passes our acceptance criteria of ±15% for each calibration point.

Also, just to confirm the prep method (derivitization) wasn't the cause, I prepared a high concentration standard and diluted that down sequentially.

Dilution curve is here: http://i43.tinypic.com/2111hc0.jpg

This is an assay specific problem, since I can run existing GCMS methods that all give perfectly linear calibration curves. This method is just proving to be a pain...

Any suggestions?

[Peter Apps]

My first tip is that if you want people to help with troubleshooting it is a good idea to provide them with full information right from the start - only now do you mention that there is a derivatization step, which opens up a host of possibilities as to what is going on, the most likely of which is an interfering impurity that the ECD sees but the MS does not.

Unless you are running this analysis on the same instrument with two detectors it is not likely that "Everything up until the point the column enters the detector is the same" - very small differences in inlet and column can generate the curves that you see.

An ECD is actually not very linear, although the depeartures from a straight line are usually masked by irreproducibility.

Looking at the curves I see three segments on the MS curve and two on the ECD curve, with kinks between them - I have seen this when making up dilutions over wide ranges as different syringes are used for different segments of the curve. I would stilll like to know whether you do point or serial dilutions.

What is the analyte and how are you derivatizing it ?, and what are your instrument conditions ? Then we can help.

Peter

[AICMM]

jdlh199,

SIM or scan? What ions?

Best regards.

[jdlh199]

Sorry for not including all the details....

The analysis is for a zinc dialkyldithiophosphate (ZDDP) in water samples, and we are derivitizing with Pentafluorobenzyl bromide (PFBB). We are creating a stock solution in ACN and then diluting that down to 10 and 1ppm. Calibration standards are prepared by mixing 4.5mL water with various quantities of the appropriate standard and ACN to make up to 5mL of solution at 10% ACN.

To the 5mL water sample, 1mL of 10mM TMAH in ACN is added, followed by 1mL 30mM PFBB in ACN. The solution is vortexed for 20 mins and then 2.5ml hexane is added for LLE. This is done twice and the 5mL hexane is evaporated to dryness and reconstituted in 0.5mL hexane.

1uL of the sample is injected onto the GC in splitless mode at 280°C, with the purge at 1min. Injector has a typical split/splitless liner with glass wool. The GC oven is programmed from 100 to 300°C at 20°/min and detection performed by either ECD or by MS by monitoring mass 353 in SIM mode. 353 is the derivitized ZDDP minus one alkyl chain (there's no sign of the molecular ion in high conc. samples under scan mode, supported by NIST MS database).

Right now I'm leaning towards an MS effect, possibly cleavage of the PFB ester during ionization that is concentration dependent effect? I'm trying some runs with different MS tune settings, lens voltages etc, to see if that makes any difference.

Thanks again for the tips!

[Don_Hilton]

What is your transfer line temperature?

[jdlh199]

250°C, although I've tried reducing that to 220°C and 180°C to no effect.

[Don_Hilton]

Try warmer. If it takes a temperature close to 300 degrees to move the compund off the column, your transfer line is a cold spot. And while compunds that take a lot of temperature to get down a column will cross a cool transfer line, heating the transfer line can help peak shape. This one I've learned the hard way.

What column and flow rate are you using?

[Peter Apps]

How does the blank look on each system ? i.e. everything except the analyte.

Peter